program(1.0) [buildInfo = dict, tensor>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}, {"coremltools-component-milinternal", ""}, {"coremltools-version", "7.1"}})] { func main(tensor encoder_hidden_states, tensor hidden_states_149_cast_fp16, tensor input_115_cast_fp16, tensor input_117_cast_fp16, tensor input_143_cast_fp16, tensor input_15_cast_fp16, tensor input_169_cast_fp16, tensor input_171_cast_fp16, tensor input_253_cast_fp16, tensor input_35_cast_fp16, tensor input_61_cast_fp16, tensor input_63_cast_fp16, tensor input_7_cast_fp16, tensor input_89_cast_fp16) { tensor cast_4_dtype_0 = const()[name = tensor("cast_4_dtype_0"), val = tensor("fp16")]; tensor add_1_mean_0_to_fp16 = const()[name = tensor("add_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; tensor add_1_variance_0_to_fp16 = const()[name = tensor("add_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768)))]; tensor cast_11_dtype_0 = const()[name = tensor("cast_11_dtype_0"), val = tensor("fp16")]; tensor cast_12_dtype_0 = const()[name = tensor("cast_12_dtype_0"), val = tensor("fp16")]; tensor cast_5_dtype_0 = const()[name = tensor("cast_5_dtype_0"), val = tensor("fp16")]; tensor cast_7_dtype_0 = const()[name = tensor("cast_7_dtype_0"), val = tensor("fp16")]; tensor add_15_mean_0_to_fp16 = const()[name = tensor("add_15_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1472)))]; tensor add_15_variance_0_to_fp16 = const()[name = tensor("add_15_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2816)))]; tensor cast_10_dtype_0 = const()[name = tensor("cast_10_dtype_0"), val = tensor("fp16")]; tensor cast_6_dtype_0 = const()[name = tensor("cast_6_dtype_0"), val = tensor("fp16")]; tensor cast_8_dtype_0 = const()[name = tensor("cast_8_dtype_0"), val = tensor("fp16")]; tensor add_27_mean_0_to_fp16 = const()[name = tensor("add_27_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4160)))]; tensor add_27_variance_0_to_fp16 = const()[name = tensor("add_27_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6784)))]; tensor cast_3_dtype_0 = const()[name = tensor("cast_3_dtype_0"), val = tensor("fp16")]; tensor cast_1_dtype_0 = const()[name = tensor("cast_1_dtype_0"), val = tensor("fp16")]; tensor cast_9_dtype_0 = const()[name = tensor("cast_9_dtype_0"), val = tensor("fp16")]; tensor var_5405 = const()[name = tensor("op_5405"), val = tensor(1)]; tensor add_55_mean_0_to_fp16 = const()[name = tensor("add_55_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9408)))]; tensor add_55_variance_0_to_fp16 = const()[name = tensor("add_55_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14592)))]; tensor cast_2_dtype_0 = const()[name = tensor("cast_2_dtype_0"), val = tensor("fp16")]; tensor cast_0_dtype_0 = const()[name = tensor("cast_0_dtype_0"), val = tensor("fp16")]; tensor var_5502 = const()[name = tensor("op_5502"), val = tensor([1, 1])]; tensor var_5504 = const()[name = tensor("op_5504"), val = tensor([1, 1])]; tensor x_7_pad_type_0 = const()[name = tensor("x_7_pad_type_0"), val = tensor("custom")]; tensor x_7_pad_0 = const()[name = tensor("x_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19776)))]; tensor up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6573440)))]; tensor cast_2 = cast(dtype = cast_2_dtype_0, x = input_253_cast_fp16)[name = tensor("cast_2")]; tensor x_7_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_5504, groups = var_5405, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = var_5502, weight = up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16, x = cast_2)[name = tensor("x_7_cast_fp16")]; tensor cast_1 = cast(dtype = cast_0_dtype_0, x = hidden_states_149_cast_fp16)[name = tensor("cast_1")]; tensor hidden_states_151_cast_fp16 = add(x = x_7_cast_fp16, y = cast_1)[name = tensor("hidden_states_151_cast_fp16")]; tensor input_267_interleave_0 = const()[name = tensor("input_267_interleave_0"), val = tensor(false)]; tensor cast_3 = cast(dtype = cast_9_dtype_0, x = input_171_cast_fp16)[name = tensor("cast_3")]; tensor input_267_cast_fp16 = concat(axis = var_5405, interleave = input_267_interleave_0, values = (hidden_states_151_cast_fp16, cast_3))[name = tensor("input_267_cast_fp16")]; tensor reshape_124_shape_0 = const()[name = tensor("reshape_124_shape_0"), val = tensor([2, 32, 80, 6, 6])]; tensor reshape_124_cast_fp16 = reshape(shape = reshape_124_shape_0, x = input_267_cast_fp16)[name = tensor("reshape_124_cast_fp16")]; tensor reduce_mean_93_axes_0 = const()[name = tensor("reduce_mean_93_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_93_keep_dims_0 = const()[name = tensor("reduce_mean_93_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_93_cast_fp16 = reduce_mean(axes = reduce_mean_93_axes_0, keep_dims = reduce_mean_93_keep_dims_0, x = reshape_124_cast_fp16)[name = tensor("reduce_mean_93_cast_fp16")]; tensor sub_62_cast_fp16 = sub(x = reshape_124_cast_fp16, y = reduce_mean_93_cast_fp16)[name = tensor("sub_62_cast_fp16")]; tensor square_31_cast_fp16 = square(x = sub_62_cast_fp16)[name = tensor("square_31_cast_fp16")]; tensor reduce_mean_95_axes_0 = const()[name = tensor("reduce_mean_95_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_95_keep_dims_0 = const()[name = tensor("reduce_mean_95_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_95_cast_fp16 = reduce_mean(axes = reduce_mean_95_axes_0, keep_dims = reduce_mean_95_keep_dims_0, x = square_31_cast_fp16)[name = tensor("reduce_mean_95_cast_fp16")]; tensor add_62_y_0_to_fp16 = const()[name = tensor("add_62_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_62_cast_fp16 = add(x = reduce_mean_95_cast_fp16, y = add_62_y_0_to_fp16)[name = tensor("add_62_cast_fp16")]; tensor sqrt_31_cast_fp16 = sqrt(x = add_62_cast_fp16)[name = tensor("sqrt_31_cast_fp16")]; tensor real_div_31_cast_fp16 = real_div(x = sub_62_cast_fp16, y = sqrt_31_cast_fp16)[name = tensor("real_div_31_cast_fp16")]; tensor reshape_125_shape_0 = const()[name = tensor("reshape_125_shape_0"), val = tensor([2, 2560, 6, 6])]; tensor reshape_125_cast_fp16 = reshape(shape = reshape_125_shape_0, x = real_div_31_cast_fp16)[name = tensor("reshape_125_cast_fp16")]; tensor add_63_gamma_0_to_fp16 = const()[name = tensor("add_63_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6576064)))]; tensor add_63_beta_0_to_fp16 = const()[name = tensor("add_63_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6581248)))]; tensor add_63_epsilon_0_to_fp16 = const()[name = tensor("add_63_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_63_cast_fp16 = batch_norm(beta = add_63_beta_0_to_fp16, epsilon = add_63_epsilon_0_to_fp16, gamma = add_63_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_125_cast_fp16)[name = tensor("add_63_cast_fp16")]; tensor input_271_cast_fp16 = silu(x = add_63_cast_fp16)[name = tensor("input_271_cast_fp16")]; tensor var_5522 = const()[name = tensor("op_5522"), val = tensor([1, 1])]; tensor var_5524 = const()[name = tensor("op_5524"), val = tensor([1, 1])]; tensor hidden_states_153_pad_type_0 = const()[name = tensor("hidden_states_153_pad_type_0"), val = tensor("custom")]; tensor hidden_states_153_pad_0 = const()[name = tensor("hidden_states_153_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_0_resnets_2_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6586432)))]; tensor up_blocks_0_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65568896)))]; tensor hidden_states_153_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv1_bias_to_fp16, dilations = var_5524, groups = var_5405, pad = hidden_states_153_pad_0, pad_type = hidden_states_153_pad_type_0, strides = var_5522, weight = up_blocks_0_resnets_2_conv1_weight_to_fp16, x = input_271_cast_fp16)[name = tensor("hidden_states_153_cast_fp16")]; tensor var_5530 = const()[name = tensor("op_5530"), val = tensor([1, 1])]; tensor var_5532 = const()[name = tensor("op_5532"), val = tensor([1, 1])]; tensor temb_25_pad_type_0 = const()[name = tensor("temb_25_pad_type_0"), val = tensor("custom")]; tensor temb_25_pad_0 = const()[name = tensor("temb_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65571520)))]; tensor up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68848384)))]; tensor cast_12 = cast(dtype = cast_11_dtype_0, x = input_15_cast_fp16)[name = tensor("cast_12")]; tensor temb_25_cast_fp16 = conv(bias = up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_5532, groups = var_5405, pad = temb_25_pad_0, pad_type = temb_25_pad_type_0, strides = var_5530, weight = up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16, x = cast_12)[name = tensor("temb_25_cast_fp16")]; tensor input_275_cast_fp16 = add(x = hidden_states_153_cast_fp16, y = temb_25_cast_fp16)[name = tensor("input_275_cast_fp16")]; tensor reshape_128_shape_0 = const()[name = tensor("reshape_128_shape_0"), val = tensor([2, 32, 40, 6, 6])]; tensor reshape_128_cast_fp16 = reshape(shape = reshape_128_shape_0, x = input_275_cast_fp16)[name = tensor("reshape_128_cast_fp16")]; tensor reduce_mean_96_axes_0 = const()[name = tensor("reduce_mean_96_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_96_keep_dims_0 = const()[name = tensor("reduce_mean_96_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_96_cast_fp16 = reduce_mean(axes = reduce_mean_96_axes_0, keep_dims = reduce_mean_96_keep_dims_0, x = reshape_128_cast_fp16)[name = tensor("reduce_mean_96_cast_fp16")]; tensor sub_64_cast_fp16 = sub(x = reshape_128_cast_fp16, y = reduce_mean_96_cast_fp16)[name = tensor("sub_64_cast_fp16")]; tensor square_32_cast_fp16 = square(x = sub_64_cast_fp16)[name = tensor("square_32_cast_fp16")]; tensor reduce_mean_98_axes_0 = const()[name = tensor("reduce_mean_98_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_98_keep_dims_0 = const()[name = tensor("reduce_mean_98_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_98_cast_fp16 = reduce_mean(axes = reduce_mean_98_axes_0, keep_dims = reduce_mean_98_keep_dims_0, x = square_32_cast_fp16)[name = tensor("reduce_mean_98_cast_fp16")]; tensor add_64_y_0_to_fp16 = const()[name = tensor("add_64_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_64_cast_fp16 = add(x = reduce_mean_98_cast_fp16, y = add_64_y_0_to_fp16)[name = tensor("add_64_cast_fp16")]; tensor sqrt_32_cast_fp16 = sqrt(x = add_64_cast_fp16)[name = tensor("sqrt_32_cast_fp16")]; tensor real_div_32_cast_fp16 = real_div(x = sub_64_cast_fp16, y = sqrt_32_cast_fp16)[name = tensor("real_div_32_cast_fp16")]; tensor reshape_129_shape_0 = const()[name = tensor("reshape_129_shape_0"), val = tensor([2, 1280, 6, 6])]; tensor reshape_129_cast_fp16 = reshape(shape = reshape_129_shape_0, x = real_div_32_cast_fp16)[name = tensor("reshape_129_cast_fp16")]; tensor add_65_gamma_0_to_fp16 = const()[name = tensor("add_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68851008)))]; tensor add_65_beta_0_to_fp16 = const()[name = tensor("add_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68853632)))]; tensor add_65_epsilon_0_to_fp16 = const()[name = tensor("add_65_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_65_cast_fp16 = batch_norm(beta = add_65_beta_0_to_fp16, epsilon = add_65_epsilon_0_to_fp16, gamma = add_65_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_129_cast_fp16)[name = tensor("add_65_cast_fp16")]; tensor input_279_cast_fp16 = silu(x = add_65_cast_fp16)[name = tensor("input_279_cast_fp16")]; tensor var_5542 = const()[name = tensor("op_5542"), val = tensor([1, 1])]; tensor var_5544 = const()[name = tensor("op_5544"), val = tensor([1, 1])]; tensor hidden_states_155_pad_type_0 = const()[name = tensor("hidden_states_155_pad_type_0"), val = tensor("custom")]; tensor hidden_states_155_pad_0 = const()[name = tensor("hidden_states_155_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_0_resnets_2_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68856256)))]; tensor up_blocks_0_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98347520)))]; tensor hidden_states_155_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv2_bias_to_fp16, dilations = var_5544, groups = var_5405, pad = hidden_states_155_pad_0, pad_type = hidden_states_155_pad_type_0, strides = var_5542, weight = up_blocks_0_resnets_2_conv2_weight_to_fp16, x = input_279_cast_fp16)[name = tensor("hidden_states_155_cast_fp16")]; tensor var_5549 = const()[name = tensor("op_5549"), val = tensor([1, 1])]; tensor var_5551 = const()[name = tensor("op_5551"), val = tensor([1, 1])]; tensor x_9_pad_type_0 = const()[name = tensor("x_9_pad_type_0"), val = tensor("custom")]; tensor x_9_pad_0 = const()[name = tensor("x_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98350144)))]; tensor up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104903808)))]; tensor x_9_cast_fp16 = conv(bias = up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_5551, groups = var_5405, pad = x_9_pad_0, pad_type = x_9_pad_type_0, strides = var_5549, weight = up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16, x = input_267_cast_fp16)[name = tensor("x_9_cast_fp16")]; tensor input_281_cast_fp16 = add(x = x_9_cast_fp16, y = hidden_states_155_cast_fp16)[name = tensor("input_281_cast_fp16")]; tensor input_283_scale_factor_height_0 = const()[name = tensor("input_283_scale_factor_height_0"), val = tensor(0x1p+1)]; tensor input_283_scale_factor_width_0 = const()[name = tensor("input_283_scale_factor_width_0"), val = tensor(0x1p+1)]; tensor input_283_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_283_scale_factor_height_0, scale_factor_width = input_283_scale_factor_width_0, x = input_281_cast_fp16)[name = tensor("input_283_cast_fp16")]; tensor var_5560 = const()[name = tensor("op_5560"), val = tensor([1, 1])]; tensor var_5562 = const()[name = tensor("op_5562"), val = tensor([1, 1])]; tensor hidden_states_157_pad_type_0 = const()[name = tensor("hidden_states_157_pad_type_0"), val = tensor("custom")]; tensor hidden_states_157_pad_0 = const()[name = tensor("hidden_states_157_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_0_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor("up_blocks_0_upsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104906432)))]; tensor up_blocks_0_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor("up_blocks_0_upsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134397696)))]; tensor hidden_states_157_cast_fp16 = conv(bias = up_blocks_0_upsamplers_0_conv_bias_to_fp16, dilations = var_5562, groups = var_5405, pad = hidden_states_157_pad_0, pad_type = hidden_states_157_pad_type_0, strides = var_5560, weight = up_blocks_0_upsamplers_0_conv_weight_to_fp16, x = input_283_cast_fp16)[name = tensor("hidden_states_157_cast_fp16")]; tensor var_5594 = const()[name = tensor("op_5594"), val = tensor(true)]; tensor var_5599 = const()[name = tensor("op_5599"), val = tensor(1)]; tensor input_285_interleave_0 = const()[name = tensor("input_285_interleave_0"), val = tensor(false)]; tensor cast_4 = cast(dtype = cast_1_dtype_0, x = input_169_cast_fp16)[name = tensor("cast_4")]; tensor input_285_cast_fp16 = concat(axis = var_5599, interleave = input_285_interleave_0, values = (hidden_states_157_cast_fp16, cast_4))[name = tensor("input_285_cast_fp16")]; tensor reshape_132_shape_0 = const()[name = tensor("reshape_132_shape_0"), val = tensor([2, 32, 80, 12, 12])]; tensor reshape_132_cast_fp16 = reshape(shape = reshape_132_shape_0, x = input_285_cast_fp16)[name = tensor("reshape_132_cast_fp16")]; tensor reduce_mean_99_axes_0 = const()[name = tensor("reduce_mean_99_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_99_keep_dims_0 = const()[name = tensor("reduce_mean_99_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_99_cast_fp16 = reduce_mean(axes = reduce_mean_99_axes_0, keep_dims = reduce_mean_99_keep_dims_0, x = reshape_132_cast_fp16)[name = tensor("reduce_mean_99_cast_fp16")]; tensor sub_66_cast_fp16 = sub(x = reshape_132_cast_fp16, y = reduce_mean_99_cast_fp16)[name = tensor("sub_66_cast_fp16")]; tensor square_33_cast_fp16 = square(x = sub_66_cast_fp16)[name = tensor("square_33_cast_fp16")]; tensor reduce_mean_101_axes_0 = const()[name = tensor("reduce_mean_101_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_101_keep_dims_0 = const()[name = tensor("reduce_mean_101_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_101_cast_fp16 = reduce_mean(axes = reduce_mean_101_axes_0, keep_dims = reduce_mean_101_keep_dims_0, x = square_33_cast_fp16)[name = tensor("reduce_mean_101_cast_fp16")]; tensor add_66_y_0_to_fp16 = const()[name = tensor("add_66_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_66_cast_fp16 = add(x = reduce_mean_101_cast_fp16, y = add_66_y_0_to_fp16)[name = tensor("add_66_cast_fp16")]; tensor sqrt_33_cast_fp16 = sqrt(x = add_66_cast_fp16)[name = tensor("sqrt_33_cast_fp16")]; tensor real_div_33_cast_fp16 = real_div(x = sub_66_cast_fp16, y = sqrt_33_cast_fp16)[name = tensor("real_div_33_cast_fp16")]; tensor reshape_133_shape_0 = const()[name = tensor("reshape_133_shape_0"), val = tensor([2, 2560, 12, 12])]; tensor reshape_133_cast_fp16 = reshape(shape = reshape_133_shape_0, x = real_div_33_cast_fp16)[name = tensor("reshape_133_cast_fp16")]; tensor add_67_gamma_0_to_fp16 = const()[name = tensor("add_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134400320)))]; tensor add_67_beta_0_to_fp16 = const()[name = tensor("add_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134405504)))]; tensor add_67_epsilon_0_to_fp16 = const()[name = tensor("add_67_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_67_cast_fp16 = batch_norm(beta = add_67_beta_0_to_fp16, epsilon = add_67_epsilon_0_to_fp16, gamma = add_67_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_133_cast_fp16)[name = tensor("add_67_cast_fp16")]; tensor input_289_cast_fp16 = silu(x = add_67_cast_fp16)[name = tensor("input_289_cast_fp16")]; tensor var_5628 = const()[name = tensor("op_5628"), val = tensor([1, 1])]; tensor var_5630 = const()[name = tensor("op_5630"), val = tensor([1, 1])]; tensor hidden_states_159_pad_type_0 = const()[name = tensor("hidden_states_159_pad_type_0"), val = tensor("custom")]; tensor hidden_states_159_pad_0 = const()[name = tensor("hidden_states_159_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_1_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134410688)))]; tensor up_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(193393152)))]; tensor hidden_states_159_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_5630, groups = var_5599, pad = hidden_states_159_pad_0, pad_type = hidden_states_159_pad_type_0, strides = var_5628, weight = up_blocks_1_resnets_0_conv1_weight_to_fp16, x = input_289_cast_fp16)[name = tensor("hidden_states_159_cast_fp16")]; tensor var_5636 = const()[name = tensor("op_5636"), val = tensor([1, 1])]; tensor var_5638 = const()[name = tensor("op_5638"), val = tensor([1, 1])]; tensor temb_27_pad_type_0 = const()[name = tensor("temb_27_pad_type_0"), val = tensor("custom")]; tensor temb_27_pad_0 = const()[name = tensor("temb_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(193395776)))]; tensor up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196672640)))]; tensor temb_27_cast_fp16 = conv(bias = up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_5638, groups = var_5599, pad = temb_27_pad_0, pad_type = temb_27_pad_type_0, strides = var_5636, weight = up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16, x = cast_12)[name = tensor("temb_27_cast_fp16")]; tensor input_293_cast_fp16 = add(x = hidden_states_159_cast_fp16, y = temb_27_cast_fp16)[name = tensor("input_293_cast_fp16")]; tensor reshape_136_shape_0 = const()[name = tensor("reshape_136_shape_0"), val = tensor([2, 32, 40, 12, 12])]; tensor reshape_136_cast_fp16 = reshape(shape = reshape_136_shape_0, x = input_293_cast_fp16)[name = tensor("reshape_136_cast_fp16")]; tensor reduce_mean_102_axes_0 = const()[name = tensor("reduce_mean_102_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_102_keep_dims_0 = const()[name = tensor("reduce_mean_102_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_102_cast_fp16 = reduce_mean(axes = reduce_mean_102_axes_0, keep_dims = reduce_mean_102_keep_dims_0, x = reshape_136_cast_fp16)[name = tensor("reduce_mean_102_cast_fp16")]; tensor sub_68_cast_fp16 = sub(x = reshape_136_cast_fp16, y = reduce_mean_102_cast_fp16)[name = tensor("sub_68_cast_fp16")]; tensor square_34_cast_fp16 = square(x = sub_68_cast_fp16)[name = tensor("square_34_cast_fp16")]; tensor reduce_mean_104_axes_0 = const()[name = tensor("reduce_mean_104_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_104_keep_dims_0 = const()[name = tensor("reduce_mean_104_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_104_cast_fp16 = reduce_mean(axes = reduce_mean_104_axes_0, keep_dims = reduce_mean_104_keep_dims_0, x = square_34_cast_fp16)[name = tensor("reduce_mean_104_cast_fp16")]; tensor add_68_y_0_to_fp16 = const()[name = tensor("add_68_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_68_cast_fp16 = add(x = reduce_mean_104_cast_fp16, y = add_68_y_0_to_fp16)[name = tensor("add_68_cast_fp16")]; tensor sqrt_34_cast_fp16 = sqrt(x = add_68_cast_fp16)[name = tensor("sqrt_34_cast_fp16")]; tensor real_div_34_cast_fp16 = real_div(x = sub_68_cast_fp16, y = sqrt_34_cast_fp16)[name = tensor("real_div_34_cast_fp16")]; tensor reshape_137_shape_0 = const()[name = tensor("reshape_137_shape_0"), val = tensor([2, 1280, 12, 12])]; tensor reshape_137_cast_fp16 = reshape(shape = reshape_137_shape_0, x = real_div_34_cast_fp16)[name = tensor("reshape_137_cast_fp16")]; tensor add_69_gamma_0_to_fp16 = const()[name = tensor("add_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196675264)))]; tensor add_69_beta_0_to_fp16 = const()[name = tensor("add_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196677888)))]; tensor add_69_epsilon_0_to_fp16 = const()[name = tensor("add_69_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_69_cast_fp16 = batch_norm(beta = add_69_beta_0_to_fp16, epsilon = add_69_epsilon_0_to_fp16, gamma = add_69_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_137_cast_fp16)[name = tensor("add_69_cast_fp16")]; tensor input_297_cast_fp16 = silu(x = add_69_cast_fp16)[name = tensor("input_297_cast_fp16")]; tensor var_5648 = const()[name = tensor("op_5648"), val = tensor([1, 1])]; tensor var_5650 = const()[name = tensor("op_5650"), val = tensor([1, 1])]; tensor hidden_states_161_pad_type_0 = const()[name = tensor("hidden_states_161_pad_type_0"), val = tensor("custom")]; tensor hidden_states_161_pad_0 = const()[name = tensor("hidden_states_161_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_1_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196680512)))]; tensor up_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226171776)))]; tensor hidden_states_161_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_5650, groups = var_5599, pad = hidden_states_161_pad_0, pad_type = hidden_states_161_pad_type_0, strides = var_5648, weight = up_blocks_1_resnets_0_conv2_weight_to_fp16, x = input_297_cast_fp16)[name = tensor("hidden_states_161_cast_fp16")]; tensor var_5655 = const()[name = tensor("op_5655"), val = tensor([1, 1])]; tensor var_5657 = const()[name = tensor("op_5657"), val = tensor([1, 1])]; tensor x_11_pad_type_0 = const()[name = tensor("x_11_pad_type_0"), val = tensor("custom")]; tensor x_11_pad_0 = const()[name = tensor("x_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226174400)))]; tensor up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(232728064)))]; tensor x_11_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_5657, groups = var_5599, pad = x_11_pad_0, pad_type = x_11_pad_type_0, strides = var_5655, weight = up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16, x = input_285_cast_fp16)[name = tensor("x_11_cast_fp16")]; tensor hidden_states_163_cast_fp16 = add(x = x_11_cast_fp16, y = hidden_states_161_cast_fp16)[name = tensor("hidden_states_163_cast_fp16")]; tensor reshape_140_shape_0 = const()[name = tensor("reshape_140_shape_0"), val = tensor([2, 32, 40, 12, 12])]; tensor reshape_140_cast_fp16 = reshape(shape = reshape_140_shape_0, x = hidden_states_163_cast_fp16)[name = tensor("reshape_140_cast_fp16")]; tensor reduce_mean_105_axes_0 = const()[name = tensor("reduce_mean_105_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_105_keep_dims_0 = const()[name = tensor("reduce_mean_105_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_105_cast_fp16 = reduce_mean(axes = reduce_mean_105_axes_0, keep_dims = reduce_mean_105_keep_dims_0, x = reshape_140_cast_fp16)[name = tensor("reduce_mean_105_cast_fp16")]; tensor sub_70_cast_fp16 = sub(x = reshape_140_cast_fp16, y = reduce_mean_105_cast_fp16)[name = tensor("sub_70_cast_fp16")]; tensor square_35_cast_fp16 = square(x = sub_70_cast_fp16)[name = tensor("square_35_cast_fp16")]; tensor reduce_mean_107_axes_0 = const()[name = tensor("reduce_mean_107_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_107_keep_dims_0 = const()[name = tensor("reduce_mean_107_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_107_cast_fp16 = reduce_mean(axes = reduce_mean_107_axes_0, keep_dims = reduce_mean_107_keep_dims_0, x = square_35_cast_fp16)[name = tensor("reduce_mean_107_cast_fp16")]; tensor add_70_y_0_to_fp16 = const()[name = tensor("add_70_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_70_cast_fp16 = add(x = reduce_mean_107_cast_fp16, y = add_70_y_0_to_fp16)[name = tensor("add_70_cast_fp16")]; tensor sqrt_35_cast_fp16 = sqrt(x = add_70_cast_fp16)[name = tensor("sqrt_35_cast_fp16")]; tensor real_div_35_cast_fp16 = real_div(x = sub_70_cast_fp16, y = sqrt_35_cast_fp16)[name = tensor("real_div_35_cast_fp16")]; tensor reshape_141_shape_0 = const()[name = tensor("reshape_141_shape_0"), val = tensor([2, 1280, 12, 12])]; tensor reshape_141_cast_fp16 = reshape(shape = reshape_141_shape_0, x = real_div_35_cast_fp16)[name = tensor("reshape_141_cast_fp16")]; tensor add_71_gamma_0_to_fp16 = const()[name = tensor("add_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(232730688)))]; tensor add_71_beta_0_to_fp16 = const()[name = tensor("add_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(232733312)))]; tensor add_71_epsilon_0_to_fp16 = const()[name = tensor("add_71_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_71_cast_fp16 = batch_norm(beta = add_71_beta_0_to_fp16, epsilon = add_71_epsilon_0_to_fp16, gamma = add_71_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_141_cast_fp16)[name = tensor("add_71_cast_fp16")]; tensor var_5677 = const()[name = tensor("op_5677"), val = tensor([1, 1])]; tensor var_5679 = const()[name = tensor("op_5679"), val = tensor([1, 1])]; tensor hidden_states_165_pad_type_0 = const()[name = tensor("hidden_states_165_pad_type_0"), val = tensor("custom")]; tensor hidden_states_165_pad_0 = const()[name = tensor("hidden_states_165_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(232735936)))]; tensor up_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236012800)))]; tensor hidden_states_165_cast_fp16 = conv(bias = up_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_5679, groups = var_5599, pad = hidden_states_165_pad_0, pad_type = hidden_states_165_pad_type_0, strides = var_5677, weight = up_blocks_1_attentions_0_proj_in_weight_to_fp16, x = add_71_cast_fp16)[name = tensor("hidden_states_165_cast_fp16")]; tensor var_5684 = const()[name = tensor("op_5684"), val = tensor([2, 1280, 1, 144])]; tensor inputs_43_cast_fp16 = reshape(shape = var_5684, x = hidden_states_165_cast_fp16)[name = tensor("inputs_43_cast_fp16")]; tensor var_5694 = const()[name = tensor("op_5694"), val = tensor([1])]; tensor channels_mean_43_cast_fp16 = reduce_mean(axes = var_5694, keep_dims = var_5594, x = inputs_43_cast_fp16)[name = tensor("channels_mean_43_cast_fp16")]; tensor zero_mean_43_cast_fp16 = sub(x = inputs_43_cast_fp16, y = channels_mean_43_cast_fp16)[name = tensor("zero_mean_43_cast_fp16")]; tensor zero_mean_sq_43_cast_fp16 = mul(x = zero_mean_43_cast_fp16, y = zero_mean_43_cast_fp16)[name = tensor("zero_mean_sq_43_cast_fp16")]; tensor var_5698 = const()[name = tensor("op_5698"), val = tensor([1])]; tensor var_5699_cast_fp16 = reduce_mean(axes = var_5698, keep_dims = var_5594, x = zero_mean_sq_43_cast_fp16)[name = tensor("op_5699_cast_fp16")]; tensor var_5700_to_fp16 = const()[name = tensor("op_5700_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_5701_cast_fp16 = add(x = var_5699_cast_fp16, y = var_5700_to_fp16)[name = tensor("op_5701_cast_fp16")]; tensor denom_43_epsilon_0_to_fp16 = const()[name = tensor("denom_43_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_43_cast_fp16 = rsqrt(epsilon = denom_43_epsilon_0_to_fp16, x = var_5701_cast_fp16)[name = tensor("denom_43_cast_fp16")]; tensor out_43_cast_fp16 = mul(x = zero_mean_43_cast_fp16, y = denom_43_cast_fp16)[name = tensor("out_43_cast_fp16")]; tensor var_5705_to_fp16 = const()[name = tensor("op_5705_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236015424)))]; tensor var_5706_cast_fp16 = add(x = out_43_cast_fp16, y = var_5705_to_fp16)[name = tensor("op_5706_cast_fp16")]; tensor var_5708_to_fp16 = const()[name = tensor("op_5708_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236018048)))]; tensor hidden_states_167_cast_fp16 = mul(x = var_5706_cast_fp16, y = var_5708_to_fp16)[name = tensor("hidden_states_167_cast_fp16")]; tensor var_5715 = const()[name = tensor("op_5715"), val = tensor([1, 1])]; tensor var_5717 = const()[name = tensor("op_5717"), val = tensor([1, 1])]; tensor q_29_pad_type_0 = const()[name = tensor("q_29_pad_type_0"), val = tensor("custom")]; tensor q_29_pad_0 = const()[name = tensor("q_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236020672)))]; tensor q_29_cast_fp16 = conv(dilations = var_5717, groups = var_5599, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = var_5715, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor("q_29_cast_fp16")]; tensor var_5721 = const()[name = tensor("op_5721"), val = tensor([1, 1])]; tensor var_5723 = const()[name = tensor("op_5723"), val = tensor([1, 1])]; tensor k_57_pad_type_0 = const()[name = tensor("k_57_pad_type_0"), val = tensor("custom")]; tensor k_57_pad_0 = const()[name = tensor("k_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239297536)))]; tensor k_57_cast_fp16 = conv(dilations = var_5723, groups = var_5599, pad = k_57_pad_0, pad_type = k_57_pad_type_0, strides = var_5721, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor("k_57_cast_fp16")]; tensor var_5727 = const()[name = tensor("op_5727"), val = tensor([1, 1])]; tensor var_5729 = const()[name = tensor("op_5729"), val = tensor([1, 1])]; tensor v_29_pad_type_0 = const()[name = tensor("v_29_pad_type_0"), val = tensor("custom")]; tensor v_29_pad_0 = const()[name = tensor("v_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242574400)))]; tensor v_29_cast_fp16 = conv(dilations = var_5729, groups = var_5599, pad = v_29_pad_0, pad_type = v_29_pad_type_0, strides = var_5727, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_167_cast_fp16)[name = tensor("v_29_cast_fp16")]; tensor var_5733_begin_0 = const()[name = tensor("op_5733_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5733_end_0 = const()[name = tensor("op_5733_end_0"), val = tensor([2, 64, 1, 144])]; tensor var_5733_end_mask_0 = const()[name = tensor("op_5733_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5733_cast_fp16 = slice_by_index(begin = var_5733_begin_0, end = var_5733_end_0, end_mask = var_5733_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5733_cast_fp16")]; tensor var_5737_begin_0 = const()[name = tensor("op_5737_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_5737_end_0 = const()[name = tensor("op_5737_end_0"), val = tensor([2, 128, 1, 144])]; tensor var_5737_end_mask_0 = const()[name = tensor("op_5737_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5737_cast_fp16 = slice_by_index(begin = var_5737_begin_0, end = var_5737_end_0, end_mask = var_5737_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5737_cast_fp16")]; tensor var_5741_begin_0 = const()[name = tensor("op_5741_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_5741_end_0 = const()[name = tensor("op_5741_end_0"), val = tensor([2, 192, 1, 144])]; tensor var_5741_end_mask_0 = const()[name = tensor("op_5741_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5741_cast_fp16 = slice_by_index(begin = var_5741_begin_0, end = var_5741_end_0, end_mask = var_5741_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5741_cast_fp16")]; tensor var_5745_begin_0 = const()[name = tensor("op_5745_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_5745_end_0 = const()[name = tensor("op_5745_end_0"), val = tensor([2, 256, 1, 144])]; tensor var_5745_end_mask_0 = const()[name = tensor("op_5745_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5745_cast_fp16 = slice_by_index(begin = var_5745_begin_0, end = var_5745_end_0, end_mask = var_5745_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5745_cast_fp16")]; tensor var_5749_begin_0 = const()[name = tensor("op_5749_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_5749_end_0 = const()[name = tensor("op_5749_end_0"), val = tensor([2, 320, 1, 144])]; tensor var_5749_end_mask_0 = const()[name = tensor("op_5749_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5749_cast_fp16 = slice_by_index(begin = var_5749_begin_0, end = var_5749_end_0, end_mask = var_5749_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5749_cast_fp16")]; tensor var_5753_begin_0 = const()[name = tensor("op_5753_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_5753_end_0 = const()[name = tensor("op_5753_end_0"), val = tensor([2, 384, 1, 144])]; tensor var_5753_end_mask_0 = const()[name = tensor("op_5753_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5753_cast_fp16 = slice_by_index(begin = var_5753_begin_0, end = var_5753_end_0, end_mask = var_5753_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5753_cast_fp16")]; tensor var_5757_begin_0 = const()[name = tensor("op_5757_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_5757_end_0 = const()[name = tensor("op_5757_end_0"), val = tensor([2, 448, 1, 144])]; tensor var_5757_end_mask_0 = const()[name = tensor("op_5757_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5757_cast_fp16 = slice_by_index(begin = var_5757_begin_0, end = var_5757_end_0, end_mask = var_5757_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5757_cast_fp16")]; tensor var_5761_begin_0 = const()[name = tensor("op_5761_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_5761_end_0 = const()[name = tensor("op_5761_end_0"), val = tensor([2, 512, 1, 144])]; tensor var_5761_end_mask_0 = const()[name = tensor("op_5761_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5761_cast_fp16 = slice_by_index(begin = var_5761_begin_0, end = var_5761_end_0, end_mask = var_5761_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5761_cast_fp16")]; tensor var_5765_begin_0 = const()[name = tensor("op_5765_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_5765_end_0 = const()[name = tensor("op_5765_end_0"), val = tensor([2, 576, 1, 144])]; tensor var_5765_end_mask_0 = const()[name = tensor("op_5765_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5765_cast_fp16 = slice_by_index(begin = var_5765_begin_0, end = var_5765_end_0, end_mask = var_5765_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5765_cast_fp16")]; tensor var_5769_begin_0 = const()[name = tensor("op_5769_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_5769_end_0 = const()[name = tensor("op_5769_end_0"), val = tensor([2, 640, 1, 144])]; tensor var_5769_end_mask_0 = const()[name = tensor("op_5769_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5769_cast_fp16 = slice_by_index(begin = var_5769_begin_0, end = var_5769_end_0, end_mask = var_5769_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5769_cast_fp16")]; tensor var_5773_begin_0 = const()[name = tensor("op_5773_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_5773_end_0 = const()[name = tensor("op_5773_end_0"), val = tensor([2, 704, 1, 144])]; tensor var_5773_end_mask_0 = const()[name = tensor("op_5773_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5773_cast_fp16 = slice_by_index(begin = var_5773_begin_0, end = var_5773_end_0, end_mask = var_5773_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5773_cast_fp16")]; tensor var_5777_begin_0 = const()[name = tensor("op_5777_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_5777_end_0 = const()[name = tensor("op_5777_end_0"), val = tensor([2, 768, 1, 144])]; tensor var_5777_end_mask_0 = const()[name = tensor("op_5777_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5777_cast_fp16 = slice_by_index(begin = var_5777_begin_0, end = var_5777_end_0, end_mask = var_5777_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5777_cast_fp16")]; tensor var_5781_begin_0 = const()[name = tensor("op_5781_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_5781_end_0 = const()[name = tensor("op_5781_end_0"), val = tensor([2, 832, 1, 144])]; tensor var_5781_end_mask_0 = const()[name = tensor("op_5781_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5781_cast_fp16 = slice_by_index(begin = var_5781_begin_0, end = var_5781_end_0, end_mask = var_5781_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5781_cast_fp16")]; tensor var_5785_begin_0 = const()[name = tensor("op_5785_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_5785_end_0 = const()[name = tensor("op_5785_end_0"), val = tensor([2, 896, 1, 144])]; tensor var_5785_end_mask_0 = const()[name = tensor("op_5785_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5785_cast_fp16 = slice_by_index(begin = var_5785_begin_0, end = var_5785_end_0, end_mask = var_5785_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5785_cast_fp16")]; tensor var_5789_begin_0 = const()[name = tensor("op_5789_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_5789_end_0 = const()[name = tensor("op_5789_end_0"), val = tensor([2, 960, 1, 144])]; tensor var_5789_end_mask_0 = const()[name = tensor("op_5789_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5789_cast_fp16 = slice_by_index(begin = var_5789_begin_0, end = var_5789_end_0, end_mask = var_5789_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5789_cast_fp16")]; tensor var_5793_begin_0 = const()[name = tensor("op_5793_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_5793_end_0 = const()[name = tensor("op_5793_end_0"), val = tensor([2, 1024, 1, 144])]; tensor var_5793_end_mask_0 = const()[name = tensor("op_5793_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5793_cast_fp16 = slice_by_index(begin = var_5793_begin_0, end = var_5793_end_0, end_mask = var_5793_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5793_cast_fp16")]; tensor var_5797_begin_0 = const()[name = tensor("op_5797_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_5797_end_0 = const()[name = tensor("op_5797_end_0"), val = tensor([2, 1088, 1, 144])]; tensor var_5797_end_mask_0 = const()[name = tensor("op_5797_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5797_cast_fp16 = slice_by_index(begin = var_5797_begin_0, end = var_5797_end_0, end_mask = var_5797_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5797_cast_fp16")]; tensor var_5801_begin_0 = const()[name = tensor("op_5801_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_5801_end_0 = const()[name = tensor("op_5801_end_0"), val = tensor([2, 1152, 1, 144])]; tensor var_5801_end_mask_0 = const()[name = tensor("op_5801_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5801_cast_fp16 = slice_by_index(begin = var_5801_begin_0, end = var_5801_end_0, end_mask = var_5801_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5801_cast_fp16")]; tensor var_5805_begin_0 = const()[name = tensor("op_5805_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_5805_end_0 = const()[name = tensor("op_5805_end_0"), val = tensor([2, 1216, 1, 144])]; tensor var_5805_end_mask_0 = const()[name = tensor("op_5805_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5805_cast_fp16 = slice_by_index(begin = var_5805_begin_0, end = var_5805_end_0, end_mask = var_5805_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5805_cast_fp16")]; tensor var_5809_begin_0 = const()[name = tensor("op_5809_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_5809_end_0 = const()[name = tensor("op_5809_end_0"), val = tensor([2, 1280, 1, 144])]; tensor var_5809_end_mask_0 = const()[name = tensor("op_5809_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5809_cast_fp16 = slice_by_index(begin = var_5809_begin_0, end = var_5809_end_0, end_mask = var_5809_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5809_cast_fp16")]; tensor k_59_perm_0 = const()[name = tensor("k_59_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_5816_begin_0 = const()[name = tensor("op_5816_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5816_end_0 = const()[name = tensor("op_5816_end_0"), val = tensor([2, 144, 1, 64])]; tensor var_5816_end_mask_0 = const()[name = tensor("op_5816_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_17 = transpose(perm = k_59_perm_0, x = k_57_cast_fp16)[name = tensor("transpose_17")]; tensor var_5816_cast_fp16 = slice_by_index(begin = var_5816_begin_0, end = var_5816_end_0, end_mask = var_5816_end_mask_0, x = transpose_17)[name = tensor("op_5816_cast_fp16")]; tensor var_5820_begin_0 = const()[name = tensor("op_5820_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_5820_end_0 = const()[name = tensor("op_5820_end_0"), val = tensor([2, 144, 1, 128])]; tensor var_5820_end_mask_0 = const()[name = tensor("op_5820_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5820_cast_fp16 = slice_by_index(begin = var_5820_begin_0, end = var_5820_end_0, end_mask = var_5820_end_mask_0, x = transpose_17)[name = tensor("op_5820_cast_fp16")]; tensor var_5824_begin_0 = const()[name = tensor("op_5824_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_5824_end_0 = const()[name = tensor("op_5824_end_0"), val = tensor([2, 144, 1, 192])]; tensor var_5824_end_mask_0 = const()[name = tensor("op_5824_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5824_cast_fp16 = slice_by_index(begin = var_5824_begin_0, end = var_5824_end_0, end_mask = var_5824_end_mask_0, x = transpose_17)[name = tensor("op_5824_cast_fp16")]; tensor var_5828_begin_0 = const()[name = tensor("op_5828_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_5828_end_0 = const()[name = tensor("op_5828_end_0"), val = tensor([2, 144, 1, 256])]; tensor var_5828_end_mask_0 = const()[name = tensor("op_5828_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5828_cast_fp16 = slice_by_index(begin = var_5828_begin_0, end = var_5828_end_0, end_mask = var_5828_end_mask_0, x = transpose_17)[name = tensor("op_5828_cast_fp16")]; tensor var_5832_begin_0 = const()[name = tensor("op_5832_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_5832_end_0 = const()[name = tensor("op_5832_end_0"), val = tensor([2, 144, 1, 320])]; tensor var_5832_end_mask_0 = const()[name = tensor("op_5832_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5832_cast_fp16 = slice_by_index(begin = var_5832_begin_0, end = var_5832_end_0, end_mask = var_5832_end_mask_0, x = transpose_17)[name = tensor("op_5832_cast_fp16")]; tensor var_5836_begin_0 = const()[name = tensor("op_5836_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_5836_end_0 = const()[name = tensor("op_5836_end_0"), val = tensor([2, 144, 1, 384])]; tensor var_5836_end_mask_0 = const()[name = tensor("op_5836_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5836_cast_fp16 = slice_by_index(begin = var_5836_begin_0, end = var_5836_end_0, end_mask = var_5836_end_mask_0, x = transpose_17)[name = tensor("op_5836_cast_fp16")]; tensor var_5840_begin_0 = const()[name = tensor("op_5840_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_5840_end_0 = const()[name = tensor("op_5840_end_0"), val = tensor([2, 144, 1, 448])]; tensor var_5840_end_mask_0 = const()[name = tensor("op_5840_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5840_cast_fp16 = slice_by_index(begin = var_5840_begin_0, end = var_5840_end_0, end_mask = var_5840_end_mask_0, x = transpose_17)[name = tensor("op_5840_cast_fp16")]; tensor var_5844_begin_0 = const()[name = tensor("op_5844_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_5844_end_0 = const()[name = tensor("op_5844_end_0"), val = tensor([2, 144, 1, 512])]; tensor var_5844_end_mask_0 = const()[name = tensor("op_5844_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5844_cast_fp16 = slice_by_index(begin = var_5844_begin_0, end = var_5844_end_0, end_mask = var_5844_end_mask_0, x = transpose_17)[name = tensor("op_5844_cast_fp16")]; tensor var_5848_begin_0 = const()[name = tensor("op_5848_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5848_end_0 = const()[name = tensor("op_5848_end_0"), val = tensor([2, 144, 1, 576])]; tensor var_5848_end_mask_0 = const()[name = tensor("op_5848_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5848_cast_fp16 = slice_by_index(begin = var_5848_begin_0, end = var_5848_end_0, end_mask = var_5848_end_mask_0, x = transpose_17)[name = tensor("op_5848_cast_fp16")]; tensor var_5852_begin_0 = const()[name = tensor("op_5852_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_5852_end_0 = const()[name = tensor("op_5852_end_0"), val = tensor([2, 144, 1, 640])]; tensor var_5852_end_mask_0 = const()[name = tensor("op_5852_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5852_cast_fp16 = slice_by_index(begin = var_5852_begin_0, end = var_5852_end_0, end_mask = var_5852_end_mask_0, x = transpose_17)[name = tensor("op_5852_cast_fp16")]; tensor var_5856_begin_0 = const()[name = tensor("op_5856_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_5856_end_0 = const()[name = tensor("op_5856_end_0"), val = tensor([2, 144, 1, 704])]; tensor var_5856_end_mask_0 = const()[name = tensor("op_5856_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5856_cast_fp16 = slice_by_index(begin = var_5856_begin_0, end = var_5856_end_0, end_mask = var_5856_end_mask_0, x = transpose_17)[name = tensor("op_5856_cast_fp16")]; tensor var_5860_begin_0 = const()[name = tensor("op_5860_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_5860_end_0 = const()[name = tensor("op_5860_end_0"), val = tensor([2, 144, 1, 768])]; tensor var_5860_end_mask_0 = const()[name = tensor("op_5860_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5860_cast_fp16 = slice_by_index(begin = var_5860_begin_0, end = var_5860_end_0, end_mask = var_5860_end_mask_0, x = transpose_17)[name = tensor("op_5860_cast_fp16")]; tensor var_5864_begin_0 = const()[name = tensor("op_5864_begin_0"), val = tensor([0, 0, 0, 768])]; tensor var_5864_end_0 = const()[name = tensor("op_5864_end_0"), val = tensor([2, 144, 1, 832])]; tensor var_5864_end_mask_0 = const()[name = tensor("op_5864_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5864_cast_fp16 = slice_by_index(begin = var_5864_begin_0, end = var_5864_end_0, end_mask = var_5864_end_mask_0, x = transpose_17)[name = tensor("op_5864_cast_fp16")]; tensor var_5868_begin_0 = const()[name = tensor("op_5868_begin_0"), val = tensor([0, 0, 0, 832])]; tensor var_5868_end_0 = const()[name = tensor("op_5868_end_0"), val = tensor([2, 144, 1, 896])]; tensor var_5868_end_mask_0 = const()[name = tensor("op_5868_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5868_cast_fp16 = slice_by_index(begin = var_5868_begin_0, end = var_5868_end_0, end_mask = var_5868_end_mask_0, x = transpose_17)[name = tensor("op_5868_cast_fp16")]; tensor var_5872_begin_0 = const()[name = tensor("op_5872_begin_0"), val = tensor([0, 0, 0, 896])]; tensor var_5872_end_0 = const()[name = tensor("op_5872_end_0"), val = tensor([2, 144, 1, 960])]; tensor var_5872_end_mask_0 = const()[name = tensor("op_5872_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5872_cast_fp16 = slice_by_index(begin = var_5872_begin_0, end = var_5872_end_0, end_mask = var_5872_end_mask_0, x = transpose_17)[name = tensor("op_5872_cast_fp16")]; tensor var_5876_begin_0 = const()[name = tensor("op_5876_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_5876_end_0 = const()[name = tensor("op_5876_end_0"), val = tensor([2, 144, 1, 1024])]; tensor var_5876_end_mask_0 = const()[name = tensor("op_5876_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5876_cast_fp16 = slice_by_index(begin = var_5876_begin_0, end = var_5876_end_0, end_mask = var_5876_end_mask_0, x = transpose_17)[name = tensor("op_5876_cast_fp16")]; tensor var_5880_begin_0 = const()[name = tensor("op_5880_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_5880_end_0 = const()[name = tensor("op_5880_end_0"), val = tensor([2, 144, 1, 1088])]; tensor var_5880_end_mask_0 = const()[name = tensor("op_5880_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5880_cast_fp16 = slice_by_index(begin = var_5880_begin_0, end = var_5880_end_0, end_mask = var_5880_end_mask_0, x = transpose_17)[name = tensor("op_5880_cast_fp16")]; tensor var_5884_begin_0 = const()[name = tensor("op_5884_begin_0"), val = tensor([0, 0, 0, 1088])]; tensor var_5884_end_0 = const()[name = tensor("op_5884_end_0"), val = tensor([2, 144, 1, 1152])]; tensor var_5884_end_mask_0 = const()[name = tensor("op_5884_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5884_cast_fp16 = slice_by_index(begin = var_5884_begin_0, end = var_5884_end_0, end_mask = var_5884_end_mask_0, x = transpose_17)[name = tensor("op_5884_cast_fp16")]; tensor var_5888_begin_0 = const()[name = tensor("op_5888_begin_0"), val = tensor([0, 0, 0, 1152])]; tensor var_5888_end_0 = const()[name = tensor("op_5888_end_0"), val = tensor([2, 144, 1, 1216])]; tensor var_5888_end_mask_0 = const()[name = tensor("op_5888_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5888_cast_fp16 = slice_by_index(begin = var_5888_begin_0, end = var_5888_end_0, end_mask = var_5888_end_mask_0, x = transpose_17)[name = tensor("op_5888_cast_fp16")]; tensor var_5892_begin_0 = const()[name = tensor("op_5892_begin_0"), val = tensor([0, 0, 0, 1216])]; tensor var_5892_end_0 = const()[name = tensor("op_5892_end_0"), val = tensor([2, 144, 1, 1280])]; tensor var_5892_end_mask_0 = const()[name = tensor("op_5892_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5892_cast_fp16 = slice_by_index(begin = var_5892_begin_0, end = var_5892_end_0, end_mask = var_5892_end_mask_0, x = transpose_17)[name = tensor("op_5892_cast_fp16")]; tensor var_5894_begin_0 = const()[name = tensor("op_5894_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5894_end_0 = const()[name = tensor("op_5894_end_0"), val = tensor([2, 64, 1, 144])]; tensor var_5894_end_mask_0 = const()[name = tensor("op_5894_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5894_cast_fp16 = slice_by_index(begin = var_5894_begin_0, end = var_5894_end_0, end_mask = var_5894_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5894_cast_fp16")]; tensor var_5898_begin_0 = const()[name = tensor("op_5898_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_5898_end_0 = const()[name = tensor("op_5898_end_0"), val = tensor([2, 128, 1, 144])]; tensor var_5898_end_mask_0 = const()[name = tensor("op_5898_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5898_cast_fp16 = slice_by_index(begin = var_5898_begin_0, end = var_5898_end_0, end_mask = var_5898_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5898_cast_fp16")]; tensor var_5902_begin_0 = const()[name = tensor("op_5902_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_5902_end_0 = const()[name = tensor("op_5902_end_0"), val = tensor([2, 192, 1, 144])]; tensor var_5902_end_mask_0 = const()[name = tensor("op_5902_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5902_cast_fp16 = slice_by_index(begin = var_5902_begin_0, end = var_5902_end_0, end_mask = var_5902_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5902_cast_fp16")]; tensor var_5906_begin_0 = const()[name = tensor("op_5906_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_5906_end_0 = const()[name = tensor("op_5906_end_0"), val = tensor([2, 256, 1, 144])]; tensor var_5906_end_mask_0 = const()[name = tensor("op_5906_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5906_cast_fp16 = slice_by_index(begin = var_5906_begin_0, end = var_5906_end_0, end_mask = var_5906_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5906_cast_fp16")]; tensor var_5910_begin_0 = const()[name = tensor("op_5910_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_5910_end_0 = const()[name = tensor("op_5910_end_0"), val = tensor([2, 320, 1, 144])]; tensor var_5910_end_mask_0 = const()[name = tensor("op_5910_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5910_cast_fp16 = slice_by_index(begin = var_5910_begin_0, end = var_5910_end_0, end_mask = var_5910_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5910_cast_fp16")]; tensor var_5914_begin_0 = const()[name = tensor("op_5914_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_5914_end_0 = const()[name = tensor("op_5914_end_0"), val = tensor([2, 384, 1, 144])]; tensor var_5914_end_mask_0 = const()[name = tensor("op_5914_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5914_cast_fp16 = slice_by_index(begin = var_5914_begin_0, end = var_5914_end_0, end_mask = var_5914_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5914_cast_fp16")]; tensor var_5918_begin_0 = const()[name = tensor("op_5918_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_5918_end_0 = const()[name = tensor("op_5918_end_0"), val = tensor([2, 448, 1, 144])]; tensor var_5918_end_mask_0 = const()[name = tensor("op_5918_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5918_cast_fp16 = slice_by_index(begin = var_5918_begin_0, end = var_5918_end_0, end_mask = var_5918_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5918_cast_fp16")]; tensor var_5922_begin_0 = const()[name = tensor("op_5922_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_5922_end_0 = const()[name = tensor("op_5922_end_0"), val = tensor([2, 512, 1, 144])]; tensor var_5922_end_mask_0 = const()[name = tensor("op_5922_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5922_cast_fp16 = slice_by_index(begin = var_5922_begin_0, end = var_5922_end_0, end_mask = var_5922_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5922_cast_fp16")]; tensor var_5926_begin_0 = const()[name = tensor("op_5926_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_5926_end_0 = const()[name = tensor("op_5926_end_0"), val = tensor([2, 576, 1, 144])]; tensor var_5926_end_mask_0 = const()[name = tensor("op_5926_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5926_cast_fp16 = slice_by_index(begin = var_5926_begin_0, end = var_5926_end_0, end_mask = var_5926_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5926_cast_fp16")]; tensor var_5930_begin_0 = const()[name = tensor("op_5930_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_5930_end_0 = const()[name = tensor("op_5930_end_0"), val = tensor([2, 640, 1, 144])]; tensor var_5930_end_mask_0 = const()[name = tensor("op_5930_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5930_cast_fp16 = slice_by_index(begin = var_5930_begin_0, end = var_5930_end_0, end_mask = var_5930_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5930_cast_fp16")]; tensor var_5934_begin_0 = const()[name = tensor("op_5934_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_5934_end_0 = const()[name = tensor("op_5934_end_0"), val = tensor([2, 704, 1, 144])]; tensor var_5934_end_mask_0 = const()[name = tensor("op_5934_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5934_cast_fp16 = slice_by_index(begin = var_5934_begin_0, end = var_5934_end_0, end_mask = var_5934_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5934_cast_fp16")]; tensor var_5938_begin_0 = const()[name = tensor("op_5938_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_5938_end_0 = const()[name = tensor("op_5938_end_0"), val = tensor([2, 768, 1, 144])]; tensor var_5938_end_mask_0 = const()[name = tensor("op_5938_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5938_cast_fp16 = slice_by_index(begin = var_5938_begin_0, end = var_5938_end_0, end_mask = var_5938_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5938_cast_fp16")]; tensor var_5942_begin_0 = const()[name = tensor("op_5942_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_5942_end_0 = const()[name = tensor("op_5942_end_0"), val = tensor([2, 832, 1, 144])]; tensor var_5942_end_mask_0 = const()[name = tensor("op_5942_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5942_cast_fp16 = slice_by_index(begin = var_5942_begin_0, end = var_5942_end_0, end_mask = var_5942_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5942_cast_fp16")]; tensor var_5946_begin_0 = const()[name = tensor("op_5946_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_5946_end_0 = const()[name = tensor("op_5946_end_0"), val = tensor([2, 896, 1, 144])]; tensor var_5946_end_mask_0 = const()[name = tensor("op_5946_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5946_cast_fp16 = slice_by_index(begin = var_5946_begin_0, end = var_5946_end_0, end_mask = var_5946_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5946_cast_fp16")]; tensor var_5950_begin_0 = const()[name = tensor("op_5950_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_5950_end_0 = const()[name = tensor("op_5950_end_0"), val = tensor([2, 960, 1, 144])]; tensor var_5950_end_mask_0 = const()[name = tensor("op_5950_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5950_cast_fp16 = slice_by_index(begin = var_5950_begin_0, end = var_5950_end_0, end_mask = var_5950_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5950_cast_fp16")]; tensor var_5954_begin_0 = const()[name = tensor("op_5954_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_5954_end_0 = const()[name = tensor("op_5954_end_0"), val = tensor([2, 1024, 1, 144])]; tensor var_5954_end_mask_0 = const()[name = tensor("op_5954_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5954_cast_fp16 = slice_by_index(begin = var_5954_begin_0, end = var_5954_end_0, end_mask = var_5954_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5954_cast_fp16")]; tensor var_5958_begin_0 = const()[name = tensor("op_5958_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_5958_end_0 = const()[name = tensor("op_5958_end_0"), val = tensor([2, 1088, 1, 144])]; tensor var_5958_end_mask_0 = const()[name = tensor("op_5958_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5958_cast_fp16 = slice_by_index(begin = var_5958_begin_0, end = var_5958_end_0, end_mask = var_5958_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5958_cast_fp16")]; tensor var_5962_begin_0 = const()[name = tensor("op_5962_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_5962_end_0 = const()[name = tensor("op_5962_end_0"), val = tensor([2, 1152, 1, 144])]; tensor var_5962_end_mask_0 = const()[name = tensor("op_5962_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5962_cast_fp16 = slice_by_index(begin = var_5962_begin_0, end = var_5962_end_0, end_mask = var_5962_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5962_cast_fp16")]; tensor var_5966_begin_0 = const()[name = tensor("op_5966_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_5966_end_0 = const()[name = tensor("op_5966_end_0"), val = tensor([2, 1216, 1, 144])]; tensor var_5966_end_mask_0 = const()[name = tensor("op_5966_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5966_cast_fp16 = slice_by_index(begin = var_5966_begin_0, end = var_5966_end_0, end_mask = var_5966_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5966_cast_fp16")]; tensor var_5970_begin_0 = const()[name = tensor("op_5970_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_5970_end_0 = const()[name = tensor("op_5970_end_0"), val = tensor([2, 1280, 1, 144])]; tensor var_5970_end_mask_0 = const()[name = tensor("op_5970_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5970_cast_fp16 = slice_by_index(begin = var_5970_begin_0, end = var_5970_end_0, end_mask = var_5970_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5970_cast_fp16")]; tensor var_5974_equation_0 = const()[name = tensor("op_5974_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5974_cast_fp16 = einsum(equation = var_5974_equation_0, values = (var_5816_cast_fp16, var_5733_cast_fp16))[name = tensor("op_5974_cast_fp16")]; tensor var_5975_to_fp16 = const()[name = tensor("op_5975_to_fp16"), val = tensor(0x1p-3)]; tensor aw_361_cast_fp16 = mul(x = var_5974_cast_fp16, y = var_5975_to_fp16)[name = tensor("aw_361_cast_fp16")]; tensor var_5978_equation_0 = const()[name = tensor("op_5978_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5978_cast_fp16 = einsum(equation = var_5978_equation_0, values = (var_5820_cast_fp16, var_5737_cast_fp16))[name = tensor("op_5978_cast_fp16")]; tensor var_5979_to_fp16 = const()[name = tensor("op_5979_to_fp16"), val = tensor(0x1p-3)]; tensor aw_363_cast_fp16 = mul(x = var_5978_cast_fp16, y = var_5979_to_fp16)[name = tensor("aw_363_cast_fp16")]; tensor var_5982_equation_0 = const()[name = tensor("op_5982_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5982_cast_fp16 = einsum(equation = var_5982_equation_0, values = (var_5824_cast_fp16, var_5741_cast_fp16))[name = tensor("op_5982_cast_fp16")]; tensor var_5983_to_fp16 = const()[name = tensor("op_5983_to_fp16"), val = tensor(0x1p-3)]; tensor aw_365_cast_fp16 = mul(x = var_5982_cast_fp16, y = var_5983_to_fp16)[name = tensor("aw_365_cast_fp16")]; tensor var_5986_equation_0 = const()[name = tensor("op_5986_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5986_cast_fp16 = einsum(equation = var_5986_equation_0, values = (var_5828_cast_fp16, var_5745_cast_fp16))[name = tensor("op_5986_cast_fp16")]; tensor var_5987_to_fp16 = const()[name = tensor("op_5987_to_fp16"), val = tensor(0x1p-3)]; tensor aw_367_cast_fp16 = mul(x = var_5986_cast_fp16, y = var_5987_to_fp16)[name = tensor("aw_367_cast_fp16")]; tensor var_5990_equation_0 = const()[name = tensor("op_5990_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5990_cast_fp16 = einsum(equation = var_5990_equation_0, values = (var_5832_cast_fp16, var_5749_cast_fp16))[name = tensor("op_5990_cast_fp16")]; tensor var_5991_to_fp16 = const()[name = tensor("op_5991_to_fp16"), val = tensor(0x1p-3)]; tensor aw_369_cast_fp16 = mul(x = var_5990_cast_fp16, y = var_5991_to_fp16)[name = tensor("aw_369_cast_fp16")]; tensor var_5994_equation_0 = const()[name = tensor("op_5994_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5994_cast_fp16 = einsum(equation = var_5994_equation_0, values = (var_5836_cast_fp16, var_5753_cast_fp16))[name = tensor("op_5994_cast_fp16")]; tensor var_5995_to_fp16 = const()[name = tensor("op_5995_to_fp16"), val = tensor(0x1p-3)]; tensor aw_371_cast_fp16 = mul(x = var_5994_cast_fp16, y = var_5995_to_fp16)[name = tensor("aw_371_cast_fp16")]; tensor var_5998_equation_0 = const()[name = tensor("op_5998_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5998_cast_fp16 = einsum(equation = var_5998_equation_0, values = (var_5840_cast_fp16, var_5757_cast_fp16))[name = tensor("op_5998_cast_fp16")]; tensor var_5999_to_fp16 = const()[name = tensor("op_5999_to_fp16"), val = tensor(0x1p-3)]; tensor aw_373_cast_fp16 = mul(x = var_5998_cast_fp16, y = var_5999_to_fp16)[name = tensor("aw_373_cast_fp16")]; tensor var_6002_equation_0 = const()[name = tensor("op_6002_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6002_cast_fp16 = einsum(equation = var_6002_equation_0, values = (var_5844_cast_fp16, var_5761_cast_fp16))[name = tensor("op_6002_cast_fp16")]; tensor var_6003_to_fp16 = const()[name = tensor("op_6003_to_fp16"), val = tensor(0x1p-3)]; tensor aw_375_cast_fp16 = mul(x = var_6002_cast_fp16, y = var_6003_to_fp16)[name = tensor("aw_375_cast_fp16")]; tensor var_6006_equation_0 = const()[name = tensor("op_6006_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6006_cast_fp16 = einsum(equation = var_6006_equation_0, values = (var_5848_cast_fp16, var_5765_cast_fp16))[name = tensor("op_6006_cast_fp16")]; tensor var_6007_to_fp16 = const()[name = tensor("op_6007_to_fp16"), val = tensor(0x1p-3)]; tensor aw_377_cast_fp16 = mul(x = var_6006_cast_fp16, y = var_6007_to_fp16)[name = tensor("aw_377_cast_fp16")]; tensor var_6010_equation_0 = const()[name = tensor("op_6010_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6010_cast_fp16 = einsum(equation = var_6010_equation_0, values = (var_5852_cast_fp16, var_5769_cast_fp16))[name = tensor("op_6010_cast_fp16")]; tensor var_6011_to_fp16 = const()[name = tensor("op_6011_to_fp16"), val = tensor(0x1p-3)]; tensor aw_379_cast_fp16 = mul(x = var_6010_cast_fp16, y = var_6011_to_fp16)[name = tensor("aw_379_cast_fp16")]; tensor var_6014_equation_0 = const()[name = tensor("op_6014_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6014_cast_fp16 = einsum(equation = var_6014_equation_0, values = (var_5856_cast_fp16, var_5773_cast_fp16))[name = tensor("op_6014_cast_fp16")]; tensor var_6015_to_fp16 = const()[name = tensor("op_6015_to_fp16"), val = tensor(0x1p-3)]; tensor aw_381_cast_fp16 = mul(x = var_6014_cast_fp16, y = var_6015_to_fp16)[name = tensor("aw_381_cast_fp16")]; tensor var_6018_equation_0 = const()[name = tensor("op_6018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6018_cast_fp16 = einsum(equation = var_6018_equation_0, values = (var_5860_cast_fp16, var_5777_cast_fp16))[name = tensor("op_6018_cast_fp16")]; tensor var_6019_to_fp16 = const()[name = tensor("op_6019_to_fp16"), val = tensor(0x1p-3)]; tensor aw_383_cast_fp16 = mul(x = var_6018_cast_fp16, y = var_6019_to_fp16)[name = tensor("aw_383_cast_fp16")]; tensor var_6022_equation_0 = const()[name = tensor("op_6022_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6022_cast_fp16 = einsum(equation = var_6022_equation_0, values = (var_5864_cast_fp16, var_5781_cast_fp16))[name = tensor("op_6022_cast_fp16")]; tensor var_6023_to_fp16 = const()[name = tensor("op_6023_to_fp16"), val = tensor(0x1p-3)]; tensor aw_385_cast_fp16 = mul(x = var_6022_cast_fp16, y = var_6023_to_fp16)[name = tensor("aw_385_cast_fp16")]; tensor var_6026_equation_0 = const()[name = tensor("op_6026_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6026_cast_fp16 = einsum(equation = var_6026_equation_0, values = (var_5868_cast_fp16, var_5785_cast_fp16))[name = tensor("op_6026_cast_fp16")]; tensor var_6027_to_fp16 = const()[name = tensor("op_6027_to_fp16"), val = tensor(0x1p-3)]; tensor aw_387_cast_fp16 = mul(x = var_6026_cast_fp16, y = var_6027_to_fp16)[name = tensor("aw_387_cast_fp16")]; tensor var_6030_equation_0 = const()[name = tensor("op_6030_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6030_cast_fp16 = einsum(equation = var_6030_equation_0, values = (var_5872_cast_fp16, var_5789_cast_fp16))[name = tensor("op_6030_cast_fp16")]; tensor var_6031_to_fp16 = const()[name = tensor("op_6031_to_fp16"), val = tensor(0x1p-3)]; tensor aw_389_cast_fp16 = mul(x = var_6030_cast_fp16, y = var_6031_to_fp16)[name = tensor("aw_389_cast_fp16")]; tensor var_6034_equation_0 = const()[name = tensor("op_6034_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6034_cast_fp16 = einsum(equation = var_6034_equation_0, values = (var_5876_cast_fp16, var_5793_cast_fp16))[name = tensor("op_6034_cast_fp16")]; tensor var_6035_to_fp16 = const()[name = tensor("op_6035_to_fp16"), val = tensor(0x1p-3)]; tensor aw_391_cast_fp16 = mul(x = var_6034_cast_fp16, y = var_6035_to_fp16)[name = tensor("aw_391_cast_fp16")]; tensor var_6038_equation_0 = const()[name = tensor("op_6038_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6038_cast_fp16 = einsum(equation = var_6038_equation_0, values = (var_5880_cast_fp16, var_5797_cast_fp16))[name = tensor("op_6038_cast_fp16")]; tensor var_6039_to_fp16 = const()[name = tensor("op_6039_to_fp16"), val = tensor(0x1p-3)]; tensor aw_393_cast_fp16 = mul(x = var_6038_cast_fp16, y = var_6039_to_fp16)[name = tensor("aw_393_cast_fp16")]; tensor var_6042_equation_0 = const()[name = tensor("op_6042_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6042_cast_fp16 = einsum(equation = var_6042_equation_0, values = (var_5884_cast_fp16, var_5801_cast_fp16))[name = tensor("op_6042_cast_fp16")]; tensor var_6043_to_fp16 = const()[name = tensor("op_6043_to_fp16"), val = tensor(0x1p-3)]; tensor aw_395_cast_fp16 = mul(x = var_6042_cast_fp16, y = var_6043_to_fp16)[name = tensor("aw_395_cast_fp16")]; tensor var_6046_equation_0 = const()[name = tensor("op_6046_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6046_cast_fp16 = einsum(equation = var_6046_equation_0, values = (var_5888_cast_fp16, var_5805_cast_fp16))[name = tensor("op_6046_cast_fp16")]; tensor var_6047_to_fp16 = const()[name = tensor("op_6047_to_fp16"), val = tensor(0x1p-3)]; tensor aw_397_cast_fp16 = mul(x = var_6046_cast_fp16, y = var_6047_to_fp16)[name = tensor("aw_397_cast_fp16")]; tensor var_6050_equation_0 = const()[name = tensor("op_6050_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6050_cast_fp16 = einsum(equation = var_6050_equation_0, values = (var_5892_cast_fp16, var_5809_cast_fp16))[name = tensor("op_6050_cast_fp16")]; tensor var_6051_to_fp16 = const()[name = tensor("op_6051_to_fp16"), val = tensor(0x1p-3)]; tensor aw_399_cast_fp16 = mul(x = var_6050_cast_fp16, y = var_6051_to_fp16)[name = tensor("aw_399_cast_fp16")]; tensor var_6053_cast_fp16 = softmax(axis = var_5599, x = aw_361_cast_fp16)[name = tensor("op_6053_cast_fp16")]; tensor var_6054_cast_fp16 = softmax(axis = var_5599, x = aw_363_cast_fp16)[name = tensor("op_6054_cast_fp16")]; tensor var_6055_cast_fp16 = softmax(axis = var_5599, x = aw_365_cast_fp16)[name = tensor("op_6055_cast_fp16")]; tensor var_6056_cast_fp16 = softmax(axis = var_5599, x = aw_367_cast_fp16)[name = tensor("op_6056_cast_fp16")]; tensor var_6057_cast_fp16 = softmax(axis = var_5599, x = aw_369_cast_fp16)[name = tensor("op_6057_cast_fp16")]; tensor var_6058_cast_fp16 = softmax(axis = var_5599, x = aw_371_cast_fp16)[name = tensor("op_6058_cast_fp16")]; tensor var_6059_cast_fp16 = softmax(axis = var_5599, x = aw_373_cast_fp16)[name = tensor("op_6059_cast_fp16")]; tensor var_6060_cast_fp16 = softmax(axis = var_5599, x = aw_375_cast_fp16)[name = tensor("op_6060_cast_fp16")]; tensor var_6061_cast_fp16 = softmax(axis = var_5599, x = aw_377_cast_fp16)[name = tensor("op_6061_cast_fp16")]; tensor var_6062_cast_fp16 = softmax(axis = var_5599, x = aw_379_cast_fp16)[name = tensor("op_6062_cast_fp16")]; tensor var_6063_cast_fp16 = softmax(axis = var_5599, x = aw_381_cast_fp16)[name = tensor("op_6063_cast_fp16")]; tensor var_6064_cast_fp16 = softmax(axis = var_5599, x = aw_383_cast_fp16)[name = tensor("op_6064_cast_fp16")]; tensor var_6065_cast_fp16 = softmax(axis = var_5599, x = aw_385_cast_fp16)[name = tensor("op_6065_cast_fp16")]; tensor var_6066_cast_fp16 = softmax(axis = var_5599, x = aw_387_cast_fp16)[name = tensor("op_6066_cast_fp16")]; tensor var_6067_cast_fp16 = softmax(axis = var_5599, x = aw_389_cast_fp16)[name = tensor("op_6067_cast_fp16")]; tensor var_6068_cast_fp16 = softmax(axis = var_5599, x = aw_391_cast_fp16)[name = tensor("op_6068_cast_fp16")]; tensor var_6069_cast_fp16 = softmax(axis = var_5599, x = aw_393_cast_fp16)[name = tensor("op_6069_cast_fp16")]; tensor var_6070_cast_fp16 = softmax(axis = var_5599, x = aw_395_cast_fp16)[name = tensor("op_6070_cast_fp16")]; tensor var_6071_cast_fp16 = softmax(axis = var_5599, x = aw_397_cast_fp16)[name = tensor("op_6071_cast_fp16")]; tensor var_6072_cast_fp16 = softmax(axis = var_5599, x = aw_399_cast_fp16)[name = tensor("op_6072_cast_fp16")]; tensor var_6074_equation_0 = const()[name = tensor("op_6074_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6074_cast_fp16 = einsum(equation = var_6074_equation_0, values = (var_5894_cast_fp16, var_6053_cast_fp16))[name = tensor("op_6074_cast_fp16")]; tensor var_6076_equation_0 = const()[name = tensor("op_6076_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6076_cast_fp16 = einsum(equation = var_6076_equation_0, values = (var_5898_cast_fp16, var_6054_cast_fp16))[name = tensor("op_6076_cast_fp16")]; tensor var_6078_equation_0 = const()[name = tensor("op_6078_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6078_cast_fp16 = einsum(equation = var_6078_equation_0, values = (var_5902_cast_fp16, var_6055_cast_fp16))[name = tensor("op_6078_cast_fp16")]; tensor var_6080_equation_0 = const()[name = tensor("op_6080_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6080_cast_fp16 = einsum(equation = var_6080_equation_0, values = (var_5906_cast_fp16, var_6056_cast_fp16))[name = tensor("op_6080_cast_fp16")]; tensor var_6082_equation_0 = const()[name = tensor("op_6082_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6082_cast_fp16 = einsum(equation = var_6082_equation_0, values = (var_5910_cast_fp16, var_6057_cast_fp16))[name = tensor("op_6082_cast_fp16")]; tensor var_6084_equation_0 = const()[name = tensor("op_6084_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6084_cast_fp16 = einsum(equation = var_6084_equation_0, values = (var_5914_cast_fp16, var_6058_cast_fp16))[name = tensor("op_6084_cast_fp16")]; tensor var_6086_equation_0 = const()[name = tensor("op_6086_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6086_cast_fp16 = einsum(equation = var_6086_equation_0, values = (var_5918_cast_fp16, var_6059_cast_fp16))[name = tensor("op_6086_cast_fp16")]; tensor var_6088_equation_0 = const()[name = tensor("op_6088_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6088_cast_fp16 = einsum(equation = var_6088_equation_0, values = (var_5922_cast_fp16, var_6060_cast_fp16))[name = tensor("op_6088_cast_fp16")]; tensor var_6090_equation_0 = const()[name = tensor("op_6090_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6090_cast_fp16 = einsum(equation = var_6090_equation_0, values = (var_5926_cast_fp16, var_6061_cast_fp16))[name = tensor("op_6090_cast_fp16")]; tensor var_6092_equation_0 = const()[name = tensor("op_6092_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6092_cast_fp16 = einsum(equation = var_6092_equation_0, values = (var_5930_cast_fp16, var_6062_cast_fp16))[name = tensor("op_6092_cast_fp16")]; tensor var_6094_equation_0 = const()[name = tensor("op_6094_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6094_cast_fp16 = einsum(equation = var_6094_equation_0, values = (var_5934_cast_fp16, var_6063_cast_fp16))[name = tensor("op_6094_cast_fp16")]; tensor var_6096_equation_0 = const()[name = tensor("op_6096_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6096_cast_fp16 = einsum(equation = var_6096_equation_0, values = (var_5938_cast_fp16, var_6064_cast_fp16))[name = tensor("op_6096_cast_fp16")]; tensor var_6098_equation_0 = const()[name = tensor("op_6098_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6098_cast_fp16 = einsum(equation = var_6098_equation_0, values = (var_5942_cast_fp16, var_6065_cast_fp16))[name = tensor("op_6098_cast_fp16")]; tensor var_6100_equation_0 = const()[name = tensor("op_6100_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6100_cast_fp16 = einsum(equation = var_6100_equation_0, values = (var_5946_cast_fp16, var_6066_cast_fp16))[name = tensor("op_6100_cast_fp16")]; tensor var_6102_equation_0 = const()[name = tensor("op_6102_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6102_cast_fp16 = einsum(equation = var_6102_equation_0, values = (var_5950_cast_fp16, var_6067_cast_fp16))[name = tensor("op_6102_cast_fp16")]; tensor var_6104_equation_0 = const()[name = tensor("op_6104_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6104_cast_fp16 = einsum(equation = var_6104_equation_0, values = (var_5954_cast_fp16, var_6068_cast_fp16))[name = tensor("op_6104_cast_fp16")]; tensor var_6106_equation_0 = const()[name = tensor("op_6106_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6106_cast_fp16 = einsum(equation = var_6106_equation_0, values = (var_5958_cast_fp16, var_6069_cast_fp16))[name = tensor("op_6106_cast_fp16")]; tensor var_6108_equation_0 = const()[name = tensor("op_6108_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6108_cast_fp16 = einsum(equation = var_6108_equation_0, values = (var_5962_cast_fp16, var_6070_cast_fp16))[name = tensor("op_6108_cast_fp16")]; tensor var_6110_equation_0 = const()[name = tensor("op_6110_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6110_cast_fp16 = einsum(equation = var_6110_equation_0, values = (var_5966_cast_fp16, var_6071_cast_fp16))[name = tensor("op_6110_cast_fp16")]; tensor var_6112_equation_0 = const()[name = tensor("op_6112_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6112_cast_fp16 = einsum(equation = var_6112_equation_0, values = (var_5970_cast_fp16, var_6072_cast_fp16))[name = tensor("op_6112_cast_fp16")]; tensor input_301_interleave_0 = const()[name = tensor("input_301_interleave_0"), val = tensor(false)]; tensor input_301_cast_fp16 = concat(axis = var_5599, interleave = input_301_interleave_0, values = (var_6074_cast_fp16, var_6076_cast_fp16, var_6078_cast_fp16, var_6080_cast_fp16, var_6082_cast_fp16, var_6084_cast_fp16, var_6086_cast_fp16, var_6088_cast_fp16, var_6090_cast_fp16, var_6092_cast_fp16, var_6094_cast_fp16, var_6096_cast_fp16, var_6098_cast_fp16, var_6100_cast_fp16, var_6102_cast_fp16, var_6104_cast_fp16, var_6106_cast_fp16, var_6108_cast_fp16, var_6110_cast_fp16, var_6112_cast_fp16))[name = tensor("input_301_cast_fp16")]; tensor var_6118 = const()[name = tensor("op_6118"), val = tensor([1, 1])]; tensor var_6120 = const()[name = tensor("op_6120"), val = tensor([1, 1])]; tensor var_6122_pad_type_0 = const()[name = tensor("op_6122_pad_type_0"), val = tensor("custom")]; tensor var_6122_pad_0 = const()[name = tensor("op_6122_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245851264)))]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249128128)))]; tensor var_6122_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_6120, groups = var_5599, pad = var_6122_pad_0, pad_type = var_6122_pad_type_0, strides = var_6118, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_301_cast_fp16)[name = tensor("op_6122_cast_fp16")]; tensor inputs_45_cast_fp16 = add(x = var_6122_cast_fp16, y = inputs_43_cast_fp16)[name = tensor("inputs_45_cast_fp16")]; tensor var_6126 = const()[name = tensor("op_6126"), val = tensor([1])]; tensor channels_mean_45_cast_fp16 = reduce_mean(axes = var_6126, keep_dims = var_5594, x = inputs_45_cast_fp16)[name = tensor("channels_mean_45_cast_fp16")]; tensor zero_mean_45_cast_fp16 = sub(x = inputs_45_cast_fp16, y = channels_mean_45_cast_fp16)[name = tensor("zero_mean_45_cast_fp16")]; tensor zero_mean_sq_45_cast_fp16 = mul(x = zero_mean_45_cast_fp16, y = zero_mean_45_cast_fp16)[name = tensor("zero_mean_sq_45_cast_fp16")]; tensor var_6130 = const()[name = tensor("op_6130"), val = tensor([1])]; tensor var_6131_cast_fp16 = reduce_mean(axes = var_6130, keep_dims = var_5594, x = zero_mean_sq_45_cast_fp16)[name = tensor("op_6131_cast_fp16")]; tensor var_6132_to_fp16 = const()[name = tensor("op_6132_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_6133_cast_fp16 = add(x = var_6131_cast_fp16, y = var_6132_to_fp16)[name = tensor("op_6133_cast_fp16")]; tensor denom_45_epsilon_0_to_fp16 = const()[name = tensor("denom_45_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_45_cast_fp16 = rsqrt(epsilon = denom_45_epsilon_0_to_fp16, x = var_6133_cast_fp16)[name = tensor("denom_45_cast_fp16")]; tensor out_45_cast_fp16 = mul(x = zero_mean_45_cast_fp16, y = denom_45_cast_fp16)[name = tensor("out_45_cast_fp16")]; tensor var_6137_to_fp16 = const()[name = tensor("op_6137_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249130752)))]; tensor var_6138_cast_fp16 = add(x = out_45_cast_fp16, y = var_6137_to_fp16)[name = tensor("op_6138_cast_fp16")]; tensor var_6140_to_fp16 = const()[name = tensor("op_6140_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249133376)))]; tensor hidden_states_169_cast_fp16 = mul(x = var_6138_cast_fp16, y = var_6140_to_fp16)[name = tensor("hidden_states_169_cast_fp16")]; tensor var_6147 = const()[name = tensor("op_6147"), val = tensor([1, 1])]; tensor var_6149 = const()[name = tensor("op_6149"), val = tensor([1, 1])]; tensor q_31_pad_type_0 = const()[name = tensor("q_31_pad_type_0"), val = tensor("custom")]; tensor q_31_pad_0 = const()[name = tensor("q_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249136000)))]; tensor q_31_cast_fp16 = conv(dilations = var_6149, groups = var_5599, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = var_6147, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_169_cast_fp16)[name = tensor("q_31_cast_fp16")]; tensor var_6153 = const()[name = tensor("op_6153"), val = tensor([1, 1])]; tensor var_6155 = const()[name = tensor("op_6155"), val = tensor([1, 1])]; tensor k_61_pad_type_0 = const()[name = tensor("k_61_pad_type_0"), val = tensor("custom")]; tensor k_61_pad_0 = const()[name = tensor("k_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252412864)))]; tensor k_61_cast_fp16 = conv(dilations = var_6155, groups = var_5599, pad = k_61_pad_0, pad_type = k_61_pad_type_0, strides = var_6153, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_61_cast_fp16")]; tensor var_6159 = const()[name = tensor("op_6159"), val = tensor([1, 1])]; tensor var_6161 = const()[name = tensor("op_6161"), val = tensor([1, 1])]; tensor v_31_pad_type_0 = const()[name = tensor("v_31_pad_type_0"), val = tensor("custom")]; tensor v_31_pad_0 = const()[name = tensor("v_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(255034368)))]; tensor v_31_cast_fp16 = conv(dilations = var_6161, groups = var_5599, pad = v_31_pad_0, pad_type = v_31_pad_type_0, strides = var_6159, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_31_cast_fp16")]; tensor var_6165_begin_0 = const()[name = tensor("op_6165_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6165_end_0 = const()[name = tensor("op_6165_end_0"), val = tensor([2, 64, 1, 144])]; tensor var_6165_end_mask_0 = const()[name = tensor("op_6165_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6165_cast_fp16 = slice_by_index(begin = var_6165_begin_0, end = var_6165_end_0, end_mask = var_6165_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6165_cast_fp16")]; tensor var_6169_begin_0 = const()[name = tensor("op_6169_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_6169_end_0 = const()[name = tensor("op_6169_end_0"), val = tensor([2, 128, 1, 144])]; tensor var_6169_end_mask_0 = const()[name = tensor("op_6169_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6169_cast_fp16 = slice_by_index(begin = var_6169_begin_0, end = var_6169_end_0, end_mask = var_6169_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6169_cast_fp16")]; tensor var_6173_begin_0 = const()[name = tensor("op_6173_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_6173_end_0 = const()[name = tensor("op_6173_end_0"), val = tensor([2, 192, 1, 144])]; tensor var_6173_end_mask_0 = const()[name = tensor("op_6173_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6173_cast_fp16 = slice_by_index(begin = var_6173_begin_0, end = var_6173_end_0, end_mask = var_6173_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6173_cast_fp16")]; tensor var_6177_begin_0 = const()[name = tensor("op_6177_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_6177_end_0 = const()[name = tensor("op_6177_end_0"), val = tensor([2, 256, 1, 144])]; tensor var_6177_end_mask_0 = const()[name = tensor("op_6177_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6177_cast_fp16 = slice_by_index(begin = var_6177_begin_0, end = var_6177_end_0, end_mask = var_6177_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6177_cast_fp16")]; tensor var_6181_begin_0 = const()[name = tensor("op_6181_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_6181_end_0 = const()[name = tensor("op_6181_end_0"), val = tensor([2, 320, 1, 144])]; tensor var_6181_end_mask_0 = const()[name = tensor("op_6181_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6181_cast_fp16 = slice_by_index(begin = var_6181_begin_0, end = var_6181_end_0, end_mask = var_6181_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6181_cast_fp16")]; tensor var_6185_begin_0 = const()[name = tensor("op_6185_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6185_end_0 = const()[name = tensor("op_6185_end_0"), val = tensor([2, 384, 1, 144])]; tensor var_6185_end_mask_0 = const()[name = tensor("op_6185_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6185_cast_fp16 = slice_by_index(begin = var_6185_begin_0, end = var_6185_end_0, end_mask = var_6185_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6185_cast_fp16")]; tensor var_6189_begin_0 = const()[name = tensor("op_6189_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_6189_end_0 = const()[name = tensor("op_6189_end_0"), val = tensor([2, 448, 1, 144])]; tensor var_6189_end_mask_0 = const()[name = tensor("op_6189_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6189_cast_fp16 = slice_by_index(begin = var_6189_begin_0, end = var_6189_end_0, end_mask = var_6189_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6189_cast_fp16")]; tensor var_6193_begin_0 = const()[name = tensor("op_6193_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_6193_end_0 = const()[name = tensor("op_6193_end_0"), val = tensor([2, 512, 1, 144])]; tensor var_6193_end_mask_0 = const()[name = tensor("op_6193_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6193_cast_fp16 = slice_by_index(begin = var_6193_begin_0, end = var_6193_end_0, end_mask = var_6193_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6193_cast_fp16")]; tensor var_6197_begin_0 = const()[name = tensor("op_6197_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_6197_end_0 = const()[name = tensor("op_6197_end_0"), val = tensor([2, 576, 1, 144])]; tensor var_6197_end_mask_0 = const()[name = tensor("op_6197_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6197_cast_fp16 = slice_by_index(begin = var_6197_begin_0, end = var_6197_end_0, end_mask = var_6197_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6197_cast_fp16")]; tensor var_6201_begin_0 = const()[name = tensor("op_6201_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_6201_end_0 = const()[name = tensor("op_6201_end_0"), val = tensor([2, 640, 1, 144])]; tensor var_6201_end_mask_0 = const()[name = tensor("op_6201_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6201_cast_fp16 = slice_by_index(begin = var_6201_begin_0, end = var_6201_end_0, end_mask = var_6201_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6201_cast_fp16")]; tensor var_6205_begin_0 = const()[name = tensor("op_6205_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6205_end_0 = const()[name = tensor("op_6205_end_0"), val = tensor([2, 704, 1, 144])]; tensor var_6205_end_mask_0 = const()[name = tensor("op_6205_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6205_cast_fp16 = slice_by_index(begin = var_6205_begin_0, end = var_6205_end_0, end_mask = var_6205_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6205_cast_fp16")]; tensor var_6209_begin_0 = const()[name = tensor("op_6209_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_6209_end_0 = const()[name = tensor("op_6209_end_0"), val = tensor([2, 768, 1, 144])]; tensor var_6209_end_mask_0 = const()[name = tensor("op_6209_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6209_cast_fp16 = slice_by_index(begin = var_6209_begin_0, end = var_6209_end_0, end_mask = var_6209_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6209_cast_fp16")]; tensor var_6213_begin_0 = const()[name = tensor("op_6213_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_6213_end_0 = const()[name = tensor("op_6213_end_0"), val = tensor([2, 832, 1, 144])]; tensor var_6213_end_mask_0 = const()[name = tensor("op_6213_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6213_cast_fp16 = slice_by_index(begin = var_6213_begin_0, end = var_6213_end_0, end_mask = var_6213_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6213_cast_fp16")]; tensor var_6217_begin_0 = const()[name = tensor("op_6217_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_6217_end_0 = const()[name = tensor("op_6217_end_0"), val = tensor([2, 896, 1, 144])]; tensor var_6217_end_mask_0 = const()[name = tensor("op_6217_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6217_cast_fp16 = slice_by_index(begin = var_6217_begin_0, end = var_6217_end_0, end_mask = var_6217_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6217_cast_fp16")]; tensor var_6221_begin_0 = const()[name = tensor("op_6221_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_6221_end_0 = const()[name = tensor("op_6221_end_0"), val = tensor([2, 960, 1, 144])]; tensor var_6221_end_mask_0 = const()[name = tensor("op_6221_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6221_cast_fp16 = slice_by_index(begin = var_6221_begin_0, end = var_6221_end_0, end_mask = var_6221_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6221_cast_fp16")]; tensor var_6225_begin_0 = const()[name = tensor("op_6225_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_6225_end_0 = const()[name = tensor("op_6225_end_0"), val = tensor([2, 1024, 1, 144])]; tensor var_6225_end_mask_0 = const()[name = tensor("op_6225_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6225_cast_fp16 = slice_by_index(begin = var_6225_begin_0, end = var_6225_end_0, end_mask = var_6225_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6225_cast_fp16")]; tensor var_6229_begin_0 = const()[name = tensor("op_6229_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_6229_end_0 = const()[name = tensor("op_6229_end_0"), val = tensor([2, 1088, 1, 144])]; tensor var_6229_end_mask_0 = const()[name = tensor("op_6229_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6229_cast_fp16 = slice_by_index(begin = var_6229_begin_0, end = var_6229_end_0, end_mask = var_6229_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6229_cast_fp16")]; tensor var_6233_begin_0 = const()[name = tensor("op_6233_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_6233_end_0 = const()[name = tensor("op_6233_end_0"), val = tensor([2, 1152, 1, 144])]; tensor var_6233_end_mask_0 = const()[name = tensor("op_6233_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6233_cast_fp16 = slice_by_index(begin = var_6233_begin_0, end = var_6233_end_0, end_mask = var_6233_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6233_cast_fp16")]; tensor var_6237_begin_0 = const()[name = tensor("op_6237_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_6237_end_0 = const()[name = tensor("op_6237_end_0"), val = tensor([2, 1216, 1, 144])]; tensor var_6237_end_mask_0 = const()[name = tensor("op_6237_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6237_cast_fp16 = slice_by_index(begin = var_6237_begin_0, end = var_6237_end_0, end_mask = var_6237_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6237_cast_fp16")]; tensor var_6241_begin_0 = const()[name = tensor("op_6241_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_6241_end_0 = const()[name = tensor("op_6241_end_0"), val = tensor([2, 1280, 1, 144])]; tensor var_6241_end_mask_0 = const()[name = tensor("op_6241_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6241_cast_fp16 = slice_by_index(begin = var_6241_begin_0, end = var_6241_end_0, end_mask = var_6241_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6241_cast_fp16")]; tensor k_63_perm_0 = const()[name = tensor("k_63_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_6248_begin_0 = const()[name = tensor("op_6248_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6248_end_0 = const()[name = tensor("op_6248_end_0"), val = tensor([2, 77, 1, 64])]; tensor var_6248_end_mask_0 = const()[name = tensor("op_6248_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_16 = transpose(perm = k_63_perm_0, x = k_61_cast_fp16)[name = tensor("transpose_16")]; tensor var_6248_cast_fp16 = slice_by_index(begin = var_6248_begin_0, end = var_6248_end_0, end_mask = var_6248_end_mask_0, x = transpose_16)[name = tensor("op_6248_cast_fp16")]; tensor var_6252_begin_0 = const()[name = tensor("op_6252_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_6252_end_0 = const()[name = tensor("op_6252_end_0"), val = tensor([2, 77, 1, 128])]; tensor var_6252_end_mask_0 = const()[name = tensor("op_6252_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6252_cast_fp16 = slice_by_index(begin = var_6252_begin_0, end = var_6252_end_0, end_mask = var_6252_end_mask_0, x = transpose_16)[name = tensor("op_6252_cast_fp16")]; tensor var_6256_begin_0 = const()[name = tensor("op_6256_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_6256_end_0 = const()[name = tensor("op_6256_end_0"), val = tensor([2, 77, 1, 192])]; tensor var_6256_end_mask_0 = const()[name = tensor("op_6256_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6256_cast_fp16 = slice_by_index(begin = var_6256_begin_0, end = var_6256_end_0, end_mask = var_6256_end_mask_0, x = transpose_16)[name = tensor("op_6256_cast_fp16")]; tensor var_6260_begin_0 = const()[name = tensor("op_6260_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_6260_end_0 = const()[name = tensor("op_6260_end_0"), val = tensor([2, 77, 1, 256])]; tensor var_6260_end_mask_0 = const()[name = tensor("op_6260_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6260_cast_fp16 = slice_by_index(begin = var_6260_begin_0, end = var_6260_end_0, end_mask = var_6260_end_mask_0, x = transpose_16)[name = tensor("op_6260_cast_fp16")]; tensor var_6264_begin_0 = const()[name = tensor("op_6264_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_6264_end_0 = const()[name = tensor("op_6264_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_6264_end_mask_0 = const()[name = tensor("op_6264_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6264_cast_fp16 = slice_by_index(begin = var_6264_begin_0, end = var_6264_end_0, end_mask = var_6264_end_mask_0, x = transpose_16)[name = tensor("op_6264_cast_fp16")]; tensor var_6268_begin_0 = const()[name = tensor("op_6268_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_6268_end_0 = const()[name = tensor("op_6268_end_0"), val = tensor([2, 77, 1, 384])]; tensor var_6268_end_mask_0 = const()[name = tensor("op_6268_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6268_cast_fp16 = slice_by_index(begin = var_6268_begin_0, end = var_6268_end_0, end_mask = var_6268_end_mask_0, x = transpose_16)[name = tensor("op_6268_cast_fp16")]; tensor var_6272_begin_0 = const()[name = tensor("op_6272_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_6272_end_0 = const()[name = tensor("op_6272_end_0"), val = tensor([2, 77, 1, 448])]; tensor var_6272_end_mask_0 = const()[name = tensor("op_6272_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6272_cast_fp16 = slice_by_index(begin = var_6272_begin_0, end = var_6272_end_0, end_mask = var_6272_end_mask_0, x = transpose_16)[name = tensor("op_6272_cast_fp16")]; tensor var_6276_begin_0 = const()[name = tensor("op_6276_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_6276_end_0 = const()[name = tensor("op_6276_end_0"), val = tensor([2, 77, 1, 512])]; tensor var_6276_end_mask_0 = const()[name = tensor("op_6276_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6276_cast_fp16 = slice_by_index(begin = var_6276_begin_0, end = var_6276_end_0, end_mask = var_6276_end_mask_0, x = transpose_16)[name = tensor("op_6276_cast_fp16")]; tensor var_6280_begin_0 = const()[name = tensor("op_6280_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6280_end_0 = const()[name = tensor("op_6280_end_0"), val = tensor([2, 77, 1, 576])]; tensor var_6280_end_mask_0 = const()[name = tensor("op_6280_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6280_cast_fp16 = slice_by_index(begin = var_6280_begin_0, end = var_6280_end_0, end_mask = var_6280_end_mask_0, x = transpose_16)[name = tensor("op_6280_cast_fp16")]; tensor var_6284_begin_0 = const()[name = tensor("op_6284_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_6284_end_0 = const()[name = tensor("op_6284_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_6284_end_mask_0 = const()[name = tensor("op_6284_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6284_cast_fp16 = slice_by_index(begin = var_6284_begin_0, end = var_6284_end_0, end_mask = var_6284_end_mask_0, x = transpose_16)[name = tensor("op_6284_cast_fp16")]; tensor var_6288_begin_0 = const()[name = tensor("op_6288_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_6288_end_0 = const()[name = tensor("op_6288_end_0"), val = tensor([2, 77, 1, 704])]; tensor var_6288_end_mask_0 = const()[name = tensor("op_6288_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6288_cast_fp16 = slice_by_index(begin = var_6288_begin_0, end = var_6288_end_0, end_mask = var_6288_end_mask_0, x = transpose_16)[name = tensor("op_6288_cast_fp16")]; tensor var_6292_begin_0 = const()[name = tensor("op_6292_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_6292_end_0 = const()[name = tensor("op_6292_end_0"), val = tensor([2, 77, 1, 768])]; tensor var_6292_end_mask_0 = const()[name = tensor("op_6292_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6292_cast_fp16 = slice_by_index(begin = var_6292_begin_0, end = var_6292_end_0, end_mask = var_6292_end_mask_0, x = transpose_16)[name = tensor("op_6292_cast_fp16")]; tensor var_6296_begin_0 = const()[name = tensor("op_6296_begin_0"), val = tensor([0, 0, 0, 768])]; tensor var_6296_end_0 = const()[name = tensor("op_6296_end_0"), val = tensor([2, 77, 1, 832])]; tensor var_6296_end_mask_0 = const()[name = tensor("op_6296_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6296_cast_fp16 = slice_by_index(begin = var_6296_begin_0, end = var_6296_end_0, end_mask = var_6296_end_mask_0, x = transpose_16)[name = tensor("op_6296_cast_fp16")]; tensor var_6300_begin_0 = const()[name = tensor("op_6300_begin_0"), val = tensor([0, 0, 0, 832])]; tensor var_6300_end_0 = const()[name = tensor("op_6300_end_0"), val = tensor([2, 77, 1, 896])]; tensor var_6300_end_mask_0 = const()[name = tensor("op_6300_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6300_cast_fp16 = slice_by_index(begin = var_6300_begin_0, end = var_6300_end_0, end_mask = var_6300_end_mask_0, x = transpose_16)[name = tensor("op_6300_cast_fp16")]; tensor var_6304_begin_0 = const()[name = tensor("op_6304_begin_0"), val = tensor([0, 0, 0, 896])]; tensor var_6304_end_0 = const()[name = tensor("op_6304_end_0"), val = tensor([2, 77, 1, 960])]; tensor var_6304_end_mask_0 = const()[name = tensor("op_6304_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6304_cast_fp16 = slice_by_index(begin = var_6304_begin_0, end = var_6304_end_0, end_mask = var_6304_end_mask_0, x = transpose_16)[name = tensor("op_6304_cast_fp16")]; tensor var_6308_begin_0 = const()[name = tensor("op_6308_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_6308_end_0 = const()[name = tensor("op_6308_end_0"), val = tensor([2, 77, 1, 1024])]; tensor var_6308_end_mask_0 = const()[name = tensor("op_6308_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6308_cast_fp16 = slice_by_index(begin = var_6308_begin_0, end = var_6308_end_0, end_mask = var_6308_end_mask_0, x = transpose_16)[name = tensor("op_6308_cast_fp16")]; tensor var_6312_begin_0 = const()[name = tensor("op_6312_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_6312_end_0 = const()[name = tensor("op_6312_end_0"), val = tensor([2, 77, 1, 1088])]; tensor var_6312_end_mask_0 = const()[name = tensor("op_6312_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6312_cast_fp16 = slice_by_index(begin = var_6312_begin_0, end = var_6312_end_0, end_mask = var_6312_end_mask_0, x = transpose_16)[name = tensor("op_6312_cast_fp16")]; tensor var_6316_begin_0 = const()[name = tensor("op_6316_begin_0"), val = tensor([0, 0, 0, 1088])]; tensor var_6316_end_0 = const()[name = tensor("op_6316_end_0"), val = tensor([2, 77, 1, 1152])]; tensor var_6316_end_mask_0 = const()[name = tensor("op_6316_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6316_cast_fp16 = slice_by_index(begin = var_6316_begin_0, end = var_6316_end_0, end_mask = var_6316_end_mask_0, x = transpose_16)[name = tensor("op_6316_cast_fp16")]; tensor var_6320_begin_0 = const()[name = tensor("op_6320_begin_0"), val = tensor([0, 0, 0, 1152])]; tensor var_6320_end_0 = const()[name = tensor("op_6320_end_0"), val = tensor([2, 77, 1, 1216])]; tensor var_6320_end_mask_0 = const()[name = tensor("op_6320_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6320_cast_fp16 = slice_by_index(begin = var_6320_begin_0, end = var_6320_end_0, end_mask = var_6320_end_mask_0, x = transpose_16)[name = tensor("op_6320_cast_fp16")]; tensor var_6324_begin_0 = const()[name = tensor("op_6324_begin_0"), val = tensor([0, 0, 0, 1216])]; tensor var_6324_end_0 = const()[name = tensor("op_6324_end_0"), val = tensor([2, 77, 1, 1280])]; tensor var_6324_end_mask_0 = const()[name = tensor("op_6324_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6324_cast_fp16 = slice_by_index(begin = var_6324_begin_0, end = var_6324_end_0, end_mask = var_6324_end_mask_0, x = transpose_16)[name = tensor("op_6324_cast_fp16")]; tensor var_6326_begin_0 = const()[name = tensor("op_6326_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6326_end_0 = const()[name = tensor("op_6326_end_0"), val = tensor([2, 64, 1, 77])]; tensor var_6326_end_mask_0 = const()[name = tensor("op_6326_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6326_cast_fp16 = slice_by_index(begin = var_6326_begin_0, end = var_6326_end_0, end_mask = var_6326_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6326_cast_fp16")]; tensor var_6330_begin_0 = const()[name = tensor("op_6330_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_6330_end_0 = const()[name = tensor("op_6330_end_0"), val = tensor([2, 128, 1, 77])]; tensor var_6330_end_mask_0 = const()[name = tensor("op_6330_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6330_cast_fp16 = slice_by_index(begin = var_6330_begin_0, end = var_6330_end_0, end_mask = var_6330_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6330_cast_fp16")]; tensor var_6334_begin_0 = const()[name = tensor("op_6334_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_6334_end_0 = const()[name = tensor("op_6334_end_0"), val = tensor([2, 192, 1, 77])]; tensor var_6334_end_mask_0 = const()[name = tensor("op_6334_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6334_cast_fp16 = slice_by_index(begin = var_6334_begin_0, end = var_6334_end_0, end_mask = var_6334_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6334_cast_fp16")]; tensor var_6338_begin_0 = const()[name = tensor("op_6338_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_6338_end_0 = const()[name = tensor("op_6338_end_0"), val = tensor([2, 256, 1, 77])]; tensor var_6338_end_mask_0 = const()[name = tensor("op_6338_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6338_cast_fp16 = slice_by_index(begin = var_6338_begin_0, end = var_6338_end_0, end_mask = var_6338_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6338_cast_fp16")]; tensor var_6342_begin_0 = const()[name = tensor("op_6342_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_6342_end_0 = const()[name = tensor("op_6342_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_6342_end_mask_0 = const()[name = tensor("op_6342_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6342_cast_fp16 = slice_by_index(begin = var_6342_begin_0, end = var_6342_end_0, end_mask = var_6342_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6342_cast_fp16")]; tensor var_6346_begin_0 = const()[name = tensor("op_6346_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6346_end_0 = const()[name = tensor("op_6346_end_0"), val = tensor([2, 384, 1, 77])]; tensor var_6346_end_mask_0 = const()[name = tensor("op_6346_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6346_cast_fp16 = slice_by_index(begin = var_6346_begin_0, end = var_6346_end_0, end_mask = var_6346_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6346_cast_fp16")]; tensor var_6350_begin_0 = const()[name = tensor("op_6350_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_6350_end_0 = const()[name = tensor("op_6350_end_0"), val = tensor([2, 448, 1, 77])]; tensor var_6350_end_mask_0 = const()[name = tensor("op_6350_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6350_cast_fp16 = slice_by_index(begin = var_6350_begin_0, end = var_6350_end_0, end_mask = var_6350_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6350_cast_fp16")]; tensor var_6354_begin_0 = const()[name = tensor("op_6354_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_6354_end_0 = const()[name = tensor("op_6354_end_0"), val = tensor([2, 512, 1, 77])]; tensor var_6354_end_mask_0 = const()[name = tensor("op_6354_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6354_cast_fp16 = slice_by_index(begin = var_6354_begin_0, end = var_6354_end_0, end_mask = var_6354_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6354_cast_fp16")]; tensor var_6358_begin_0 = const()[name = tensor("op_6358_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_6358_end_0 = const()[name = tensor("op_6358_end_0"), val = tensor([2, 576, 1, 77])]; tensor var_6358_end_mask_0 = const()[name = tensor("op_6358_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6358_cast_fp16 = slice_by_index(begin = var_6358_begin_0, end = var_6358_end_0, end_mask = var_6358_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6358_cast_fp16")]; tensor var_6362_begin_0 = const()[name = tensor("op_6362_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_6362_end_0 = const()[name = tensor("op_6362_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_6362_end_mask_0 = const()[name = tensor("op_6362_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6362_cast_fp16 = slice_by_index(begin = var_6362_begin_0, end = var_6362_end_0, end_mask = var_6362_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6362_cast_fp16")]; tensor var_6366_begin_0 = const()[name = tensor("op_6366_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6366_end_0 = const()[name = tensor("op_6366_end_0"), val = tensor([2, 704, 1, 77])]; tensor var_6366_end_mask_0 = const()[name = tensor("op_6366_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6366_cast_fp16 = slice_by_index(begin = var_6366_begin_0, end = var_6366_end_0, end_mask = var_6366_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6366_cast_fp16")]; tensor var_6370_begin_0 = const()[name = tensor("op_6370_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_6370_end_0 = const()[name = tensor("op_6370_end_0"), val = tensor([2, 768, 1, 77])]; tensor var_6370_end_mask_0 = const()[name = tensor("op_6370_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6370_cast_fp16 = slice_by_index(begin = var_6370_begin_0, end = var_6370_end_0, end_mask = var_6370_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6370_cast_fp16")]; tensor var_6374_begin_0 = const()[name = tensor("op_6374_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_6374_end_0 = const()[name = tensor("op_6374_end_0"), val = tensor([2, 832, 1, 77])]; tensor var_6374_end_mask_0 = const()[name = tensor("op_6374_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6374_cast_fp16 = slice_by_index(begin = var_6374_begin_0, end = var_6374_end_0, end_mask = var_6374_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6374_cast_fp16")]; tensor var_6378_begin_0 = const()[name = tensor("op_6378_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_6378_end_0 = const()[name = tensor("op_6378_end_0"), val = tensor([2, 896, 1, 77])]; tensor var_6378_end_mask_0 = const()[name = tensor("op_6378_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6378_cast_fp16 = slice_by_index(begin = var_6378_begin_0, end = var_6378_end_0, end_mask = var_6378_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6378_cast_fp16")]; tensor var_6382_begin_0 = const()[name = tensor("op_6382_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_6382_end_0 = const()[name = tensor("op_6382_end_0"), val = tensor([2, 960, 1, 77])]; tensor var_6382_end_mask_0 = const()[name = tensor("op_6382_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6382_cast_fp16 = slice_by_index(begin = var_6382_begin_0, end = var_6382_end_0, end_mask = var_6382_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6382_cast_fp16")]; tensor var_6386_begin_0 = const()[name = tensor("op_6386_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_6386_end_0 = const()[name = tensor("op_6386_end_0"), val = tensor([2, 1024, 1, 77])]; tensor var_6386_end_mask_0 = const()[name = tensor("op_6386_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6386_cast_fp16 = slice_by_index(begin = var_6386_begin_0, end = var_6386_end_0, end_mask = var_6386_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6386_cast_fp16")]; tensor var_6390_begin_0 = const()[name = tensor("op_6390_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_6390_end_0 = const()[name = tensor("op_6390_end_0"), val = tensor([2, 1088, 1, 77])]; tensor var_6390_end_mask_0 = const()[name = tensor("op_6390_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6390_cast_fp16 = slice_by_index(begin = var_6390_begin_0, end = var_6390_end_0, end_mask = var_6390_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6390_cast_fp16")]; tensor var_6394_begin_0 = const()[name = tensor("op_6394_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_6394_end_0 = const()[name = tensor("op_6394_end_0"), val = tensor([2, 1152, 1, 77])]; tensor var_6394_end_mask_0 = const()[name = tensor("op_6394_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6394_cast_fp16 = slice_by_index(begin = var_6394_begin_0, end = var_6394_end_0, end_mask = var_6394_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6394_cast_fp16")]; tensor var_6398_begin_0 = const()[name = tensor("op_6398_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_6398_end_0 = const()[name = tensor("op_6398_end_0"), val = tensor([2, 1216, 1, 77])]; tensor var_6398_end_mask_0 = const()[name = tensor("op_6398_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6398_cast_fp16 = slice_by_index(begin = var_6398_begin_0, end = var_6398_end_0, end_mask = var_6398_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6398_cast_fp16")]; tensor var_6402_begin_0 = const()[name = tensor("op_6402_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_6402_end_0 = const()[name = tensor("op_6402_end_0"), val = tensor([2, 1280, 1, 77])]; tensor var_6402_end_mask_0 = const()[name = tensor("op_6402_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6402_cast_fp16 = slice_by_index(begin = var_6402_begin_0, end = var_6402_end_0, end_mask = var_6402_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6402_cast_fp16")]; tensor var_6406_equation_0 = const()[name = tensor("op_6406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6406_cast_fp16 = einsum(equation = var_6406_equation_0, values = (var_6248_cast_fp16, var_6165_cast_fp16))[name = tensor("op_6406_cast_fp16")]; tensor var_6407_to_fp16 = const()[name = tensor("op_6407_to_fp16"), val = tensor(0x1p-3)]; tensor aw_401_cast_fp16 = mul(x = var_6406_cast_fp16, y = var_6407_to_fp16)[name = tensor("aw_401_cast_fp16")]; tensor var_6410_equation_0 = const()[name = tensor("op_6410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6410_cast_fp16 = einsum(equation = var_6410_equation_0, values = (var_6252_cast_fp16, var_6169_cast_fp16))[name = tensor("op_6410_cast_fp16")]; tensor var_6411_to_fp16 = const()[name = tensor("op_6411_to_fp16"), val = tensor(0x1p-3)]; tensor aw_403_cast_fp16 = mul(x = var_6410_cast_fp16, y = var_6411_to_fp16)[name = tensor("aw_403_cast_fp16")]; tensor var_6414_equation_0 = const()[name = tensor("op_6414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6414_cast_fp16 = einsum(equation = var_6414_equation_0, values = (var_6256_cast_fp16, var_6173_cast_fp16))[name = tensor("op_6414_cast_fp16")]; tensor var_6415_to_fp16 = const()[name = tensor("op_6415_to_fp16"), val = tensor(0x1p-3)]; tensor aw_405_cast_fp16 = mul(x = var_6414_cast_fp16, y = var_6415_to_fp16)[name = tensor("aw_405_cast_fp16")]; tensor var_6418_equation_0 = const()[name = tensor("op_6418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6418_cast_fp16 = einsum(equation = var_6418_equation_0, values = (var_6260_cast_fp16, var_6177_cast_fp16))[name = tensor("op_6418_cast_fp16")]; tensor var_6419_to_fp16 = const()[name = tensor("op_6419_to_fp16"), val = tensor(0x1p-3)]; tensor aw_407_cast_fp16 = mul(x = var_6418_cast_fp16, y = var_6419_to_fp16)[name = tensor("aw_407_cast_fp16")]; tensor var_6422_equation_0 = const()[name = tensor("op_6422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6422_cast_fp16 = einsum(equation = var_6422_equation_0, values = (var_6264_cast_fp16, var_6181_cast_fp16))[name = tensor("op_6422_cast_fp16")]; tensor var_6423_to_fp16 = const()[name = tensor("op_6423_to_fp16"), val = tensor(0x1p-3)]; tensor aw_409_cast_fp16 = mul(x = var_6422_cast_fp16, y = var_6423_to_fp16)[name = tensor("aw_409_cast_fp16")]; tensor var_6426_equation_0 = const()[name = tensor("op_6426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6426_cast_fp16 = einsum(equation = var_6426_equation_0, values = (var_6268_cast_fp16, var_6185_cast_fp16))[name = tensor("op_6426_cast_fp16")]; tensor var_6427_to_fp16 = const()[name = tensor("op_6427_to_fp16"), val = tensor(0x1p-3)]; tensor aw_411_cast_fp16 = mul(x = var_6426_cast_fp16, y = var_6427_to_fp16)[name = tensor("aw_411_cast_fp16")]; tensor var_6430_equation_0 = const()[name = tensor("op_6430_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6430_cast_fp16 = einsum(equation = var_6430_equation_0, values = (var_6272_cast_fp16, var_6189_cast_fp16))[name = tensor("op_6430_cast_fp16")]; tensor var_6431_to_fp16 = const()[name = tensor("op_6431_to_fp16"), val = tensor(0x1p-3)]; tensor aw_413_cast_fp16 = mul(x = var_6430_cast_fp16, y = var_6431_to_fp16)[name = tensor("aw_413_cast_fp16")]; tensor var_6434_equation_0 = const()[name = tensor("op_6434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6434_cast_fp16 = einsum(equation = var_6434_equation_0, values = (var_6276_cast_fp16, var_6193_cast_fp16))[name = tensor("op_6434_cast_fp16")]; tensor var_6435_to_fp16 = const()[name = tensor("op_6435_to_fp16"), val = tensor(0x1p-3)]; tensor aw_415_cast_fp16 = mul(x = var_6434_cast_fp16, y = var_6435_to_fp16)[name = tensor("aw_415_cast_fp16")]; tensor var_6438_equation_0 = const()[name = tensor("op_6438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6438_cast_fp16 = einsum(equation = var_6438_equation_0, values = (var_6280_cast_fp16, var_6197_cast_fp16))[name = tensor("op_6438_cast_fp16")]; tensor var_6439_to_fp16 = const()[name = tensor("op_6439_to_fp16"), val = tensor(0x1p-3)]; tensor aw_417_cast_fp16 = mul(x = var_6438_cast_fp16, y = var_6439_to_fp16)[name = tensor("aw_417_cast_fp16")]; tensor var_6442_equation_0 = const()[name = tensor("op_6442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6442_cast_fp16 = einsum(equation = var_6442_equation_0, values = (var_6284_cast_fp16, var_6201_cast_fp16))[name = tensor("op_6442_cast_fp16")]; tensor var_6443_to_fp16 = const()[name = tensor("op_6443_to_fp16"), val = tensor(0x1p-3)]; tensor aw_419_cast_fp16 = mul(x = var_6442_cast_fp16, y = var_6443_to_fp16)[name = tensor("aw_419_cast_fp16")]; tensor var_6446_equation_0 = const()[name = tensor("op_6446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6446_cast_fp16 = einsum(equation = var_6446_equation_0, values = (var_6288_cast_fp16, var_6205_cast_fp16))[name = tensor("op_6446_cast_fp16")]; tensor var_6447_to_fp16 = const()[name = tensor("op_6447_to_fp16"), val = tensor(0x1p-3)]; tensor aw_421_cast_fp16 = mul(x = var_6446_cast_fp16, y = var_6447_to_fp16)[name = tensor("aw_421_cast_fp16")]; tensor var_6450_equation_0 = const()[name = tensor("op_6450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6450_cast_fp16 = einsum(equation = var_6450_equation_0, values = (var_6292_cast_fp16, var_6209_cast_fp16))[name = tensor("op_6450_cast_fp16")]; tensor var_6451_to_fp16 = const()[name = tensor("op_6451_to_fp16"), val = tensor(0x1p-3)]; tensor aw_423_cast_fp16 = mul(x = var_6450_cast_fp16, y = var_6451_to_fp16)[name = tensor("aw_423_cast_fp16")]; tensor var_6454_equation_0 = const()[name = tensor("op_6454_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6454_cast_fp16 = einsum(equation = var_6454_equation_0, values = (var_6296_cast_fp16, var_6213_cast_fp16))[name = tensor("op_6454_cast_fp16")]; tensor var_6455_to_fp16 = const()[name = tensor("op_6455_to_fp16"), val = tensor(0x1p-3)]; tensor aw_425_cast_fp16 = mul(x = var_6454_cast_fp16, y = var_6455_to_fp16)[name = tensor("aw_425_cast_fp16")]; tensor var_6458_equation_0 = const()[name = tensor("op_6458_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6458_cast_fp16 = einsum(equation = var_6458_equation_0, values = (var_6300_cast_fp16, var_6217_cast_fp16))[name = tensor("op_6458_cast_fp16")]; tensor var_6459_to_fp16 = const()[name = tensor("op_6459_to_fp16"), val = tensor(0x1p-3)]; tensor aw_427_cast_fp16 = mul(x = var_6458_cast_fp16, y = var_6459_to_fp16)[name = tensor("aw_427_cast_fp16")]; tensor var_6462_equation_0 = const()[name = tensor("op_6462_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6462_cast_fp16 = einsum(equation = var_6462_equation_0, values = (var_6304_cast_fp16, var_6221_cast_fp16))[name = tensor("op_6462_cast_fp16")]; tensor var_6463_to_fp16 = const()[name = tensor("op_6463_to_fp16"), val = tensor(0x1p-3)]; tensor aw_429_cast_fp16 = mul(x = var_6462_cast_fp16, y = var_6463_to_fp16)[name = tensor("aw_429_cast_fp16")]; tensor var_6466_equation_0 = const()[name = tensor("op_6466_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6466_cast_fp16 = einsum(equation = var_6466_equation_0, values = (var_6308_cast_fp16, var_6225_cast_fp16))[name = tensor("op_6466_cast_fp16")]; tensor var_6467_to_fp16 = const()[name = tensor("op_6467_to_fp16"), val = tensor(0x1p-3)]; tensor aw_431_cast_fp16 = mul(x = var_6466_cast_fp16, y = var_6467_to_fp16)[name = tensor("aw_431_cast_fp16")]; tensor var_6470_equation_0 = const()[name = tensor("op_6470_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6470_cast_fp16 = einsum(equation = var_6470_equation_0, values = (var_6312_cast_fp16, var_6229_cast_fp16))[name = tensor("op_6470_cast_fp16")]; tensor var_6471_to_fp16 = const()[name = tensor("op_6471_to_fp16"), val = tensor(0x1p-3)]; tensor aw_433_cast_fp16 = mul(x = var_6470_cast_fp16, y = var_6471_to_fp16)[name = tensor("aw_433_cast_fp16")]; tensor var_6474_equation_0 = const()[name = tensor("op_6474_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6474_cast_fp16 = einsum(equation = var_6474_equation_0, values = (var_6316_cast_fp16, var_6233_cast_fp16))[name = tensor("op_6474_cast_fp16")]; tensor var_6475_to_fp16 = const()[name = tensor("op_6475_to_fp16"), val = tensor(0x1p-3)]; tensor aw_435_cast_fp16 = mul(x = var_6474_cast_fp16, y = var_6475_to_fp16)[name = tensor("aw_435_cast_fp16")]; tensor var_6478_equation_0 = const()[name = tensor("op_6478_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6478_cast_fp16 = einsum(equation = var_6478_equation_0, values = (var_6320_cast_fp16, var_6237_cast_fp16))[name = tensor("op_6478_cast_fp16")]; tensor var_6479_to_fp16 = const()[name = tensor("op_6479_to_fp16"), val = tensor(0x1p-3)]; tensor aw_437_cast_fp16 = mul(x = var_6478_cast_fp16, y = var_6479_to_fp16)[name = tensor("aw_437_cast_fp16")]; tensor var_6482_equation_0 = const()[name = tensor("op_6482_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6482_cast_fp16 = einsum(equation = var_6482_equation_0, values = (var_6324_cast_fp16, var_6241_cast_fp16))[name = tensor("op_6482_cast_fp16")]; tensor var_6483_to_fp16 = const()[name = tensor("op_6483_to_fp16"), val = tensor(0x1p-3)]; tensor aw_439_cast_fp16 = mul(x = var_6482_cast_fp16, y = var_6483_to_fp16)[name = tensor("aw_439_cast_fp16")]; tensor var_6485_cast_fp16 = softmax(axis = var_5599, x = aw_401_cast_fp16)[name = tensor("op_6485_cast_fp16")]; tensor var_6486_cast_fp16 = softmax(axis = var_5599, x = aw_403_cast_fp16)[name = tensor("op_6486_cast_fp16")]; tensor var_6487_cast_fp16 = softmax(axis = var_5599, x = aw_405_cast_fp16)[name = tensor("op_6487_cast_fp16")]; tensor var_6488_cast_fp16 = softmax(axis = var_5599, x = aw_407_cast_fp16)[name = tensor("op_6488_cast_fp16")]; tensor var_6489_cast_fp16 = softmax(axis = var_5599, x = aw_409_cast_fp16)[name = tensor("op_6489_cast_fp16")]; tensor var_6490_cast_fp16 = softmax(axis = var_5599, x = aw_411_cast_fp16)[name = tensor("op_6490_cast_fp16")]; tensor var_6491_cast_fp16 = softmax(axis = var_5599, x = aw_413_cast_fp16)[name = tensor("op_6491_cast_fp16")]; tensor var_6492_cast_fp16 = softmax(axis = var_5599, x = aw_415_cast_fp16)[name = tensor("op_6492_cast_fp16")]; tensor var_6493_cast_fp16 = softmax(axis = var_5599, x = aw_417_cast_fp16)[name = tensor("op_6493_cast_fp16")]; tensor var_6494_cast_fp16 = softmax(axis = var_5599, x = aw_419_cast_fp16)[name = tensor("op_6494_cast_fp16")]; tensor var_6495_cast_fp16 = softmax(axis = var_5599, x = aw_421_cast_fp16)[name = tensor("op_6495_cast_fp16")]; tensor var_6496_cast_fp16 = softmax(axis = var_5599, x = aw_423_cast_fp16)[name = tensor("op_6496_cast_fp16")]; tensor var_6497_cast_fp16 = softmax(axis = var_5599, x = aw_425_cast_fp16)[name = tensor("op_6497_cast_fp16")]; tensor var_6498_cast_fp16 = softmax(axis = var_5599, x = aw_427_cast_fp16)[name = tensor("op_6498_cast_fp16")]; tensor var_6499_cast_fp16 = softmax(axis = var_5599, x = aw_429_cast_fp16)[name = tensor("op_6499_cast_fp16")]; tensor var_6500_cast_fp16 = softmax(axis = var_5599, x = aw_431_cast_fp16)[name = tensor("op_6500_cast_fp16")]; tensor var_6501_cast_fp16 = softmax(axis = var_5599, x = aw_433_cast_fp16)[name = tensor("op_6501_cast_fp16")]; tensor var_6502_cast_fp16 = softmax(axis = var_5599, x = aw_435_cast_fp16)[name = tensor("op_6502_cast_fp16")]; tensor var_6503_cast_fp16 = softmax(axis = var_5599, x = aw_437_cast_fp16)[name = tensor("op_6503_cast_fp16")]; tensor var_6504_cast_fp16 = softmax(axis = var_5599, x = aw_439_cast_fp16)[name = tensor("op_6504_cast_fp16")]; tensor var_6506_equation_0 = const()[name = tensor("op_6506_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6506_cast_fp16 = einsum(equation = var_6506_equation_0, values = (var_6326_cast_fp16, var_6485_cast_fp16))[name = tensor("op_6506_cast_fp16")]; tensor var_6508_equation_0 = const()[name = tensor("op_6508_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6508_cast_fp16 = einsum(equation = var_6508_equation_0, values = (var_6330_cast_fp16, var_6486_cast_fp16))[name = tensor("op_6508_cast_fp16")]; tensor var_6510_equation_0 = const()[name = tensor("op_6510_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6510_cast_fp16 = einsum(equation = var_6510_equation_0, values = (var_6334_cast_fp16, var_6487_cast_fp16))[name = tensor("op_6510_cast_fp16")]; tensor var_6512_equation_0 = const()[name = tensor("op_6512_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6512_cast_fp16 = einsum(equation = var_6512_equation_0, values = (var_6338_cast_fp16, var_6488_cast_fp16))[name = tensor("op_6512_cast_fp16")]; tensor var_6514_equation_0 = const()[name = tensor("op_6514_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6514_cast_fp16 = einsum(equation = var_6514_equation_0, values = (var_6342_cast_fp16, var_6489_cast_fp16))[name = tensor("op_6514_cast_fp16")]; tensor var_6516_equation_0 = const()[name = tensor("op_6516_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6516_cast_fp16 = einsum(equation = var_6516_equation_0, values = (var_6346_cast_fp16, var_6490_cast_fp16))[name = tensor("op_6516_cast_fp16")]; tensor var_6518_equation_0 = const()[name = tensor("op_6518_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6518_cast_fp16 = einsum(equation = var_6518_equation_0, values = (var_6350_cast_fp16, var_6491_cast_fp16))[name = tensor("op_6518_cast_fp16")]; tensor var_6520_equation_0 = const()[name = tensor("op_6520_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6520_cast_fp16 = einsum(equation = var_6520_equation_0, values = (var_6354_cast_fp16, var_6492_cast_fp16))[name = tensor("op_6520_cast_fp16")]; tensor var_6522_equation_0 = const()[name = tensor("op_6522_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6522_cast_fp16 = einsum(equation = var_6522_equation_0, values = (var_6358_cast_fp16, var_6493_cast_fp16))[name = tensor("op_6522_cast_fp16")]; tensor var_6524_equation_0 = const()[name = tensor("op_6524_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6524_cast_fp16 = einsum(equation = var_6524_equation_0, values = (var_6362_cast_fp16, var_6494_cast_fp16))[name = tensor("op_6524_cast_fp16")]; tensor var_6526_equation_0 = const()[name = tensor("op_6526_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6526_cast_fp16 = einsum(equation = var_6526_equation_0, values = (var_6366_cast_fp16, var_6495_cast_fp16))[name = tensor("op_6526_cast_fp16")]; tensor var_6528_equation_0 = const()[name = tensor("op_6528_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6528_cast_fp16 = einsum(equation = var_6528_equation_0, values = (var_6370_cast_fp16, var_6496_cast_fp16))[name = tensor("op_6528_cast_fp16")]; tensor var_6530_equation_0 = const()[name = tensor("op_6530_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6530_cast_fp16 = einsum(equation = var_6530_equation_0, values = (var_6374_cast_fp16, var_6497_cast_fp16))[name = tensor("op_6530_cast_fp16")]; tensor var_6532_equation_0 = const()[name = tensor("op_6532_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6532_cast_fp16 = einsum(equation = var_6532_equation_0, values = (var_6378_cast_fp16, var_6498_cast_fp16))[name = tensor("op_6532_cast_fp16")]; tensor var_6534_equation_0 = const()[name = tensor("op_6534_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6534_cast_fp16 = einsum(equation = var_6534_equation_0, values = (var_6382_cast_fp16, var_6499_cast_fp16))[name = tensor("op_6534_cast_fp16")]; tensor var_6536_equation_0 = const()[name = tensor("op_6536_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6536_cast_fp16 = einsum(equation = var_6536_equation_0, values = (var_6386_cast_fp16, var_6500_cast_fp16))[name = tensor("op_6536_cast_fp16")]; tensor var_6538_equation_0 = const()[name = tensor("op_6538_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6538_cast_fp16 = einsum(equation = var_6538_equation_0, values = (var_6390_cast_fp16, var_6501_cast_fp16))[name = tensor("op_6538_cast_fp16")]; tensor var_6540_equation_0 = const()[name = tensor("op_6540_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6540_cast_fp16 = einsum(equation = var_6540_equation_0, values = (var_6394_cast_fp16, var_6502_cast_fp16))[name = tensor("op_6540_cast_fp16")]; tensor var_6542_equation_0 = const()[name = tensor("op_6542_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6542_cast_fp16 = einsum(equation = var_6542_equation_0, values = (var_6398_cast_fp16, var_6503_cast_fp16))[name = tensor("op_6542_cast_fp16")]; tensor var_6544_equation_0 = const()[name = tensor("op_6544_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6544_cast_fp16 = einsum(equation = var_6544_equation_0, values = (var_6402_cast_fp16, var_6504_cast_fp16))[name = tensor("op_6544_cast_fp16")]; tensor input_303_interleave_0 = const()[name = tensor("input_303_interleave_0"), val = tensor(false)]; tensor input_303_cast_fp16 = concat(axis = var_5599, interleave = input_303_interleave_0, values = (var_6506_cast_fp16, var_6508_cast_fp16, var_6510_cast_fp16, var_6512_cast_fp16, var_6514_cast_fp16, var_6516_cast_fp16, var_6518_cast_fp16, var_6520_cast_fp16, var_6522_cast_fp16, var_6524_cast_fp16, var_6526_cast_fp16, var_6528_cast_fp16, var_6530_cast_fp16, var_6532_cast_fp16, var_6534_cast_fp16, var_6536_cast_fp16, var_6538_cast_fp16, var_6540_cast_fp16, var_6542_cast_fp16, var_6544_cast_fp16))[name = tensor("input_303_cast_fp16")]; tensor var_6550 = const()[name = tensor("op_6550"), val = tensor([1, 1])]; tensor var_6552 = const()[name = tensor("op_6552"), val = tensor([1, 1])]; tensor var_6554_pad_type_0 = const()[name = tensor("op_6554_pad_type_0"), val = tensor("custom")]; tensor var_6554_pad_0 = const()[name = tensor("op_6554_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257655872)))]; tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260932736)))]; tensor var_6554_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_6552, groups = var_5599, pad = var_6554_pad_0, pad_type = var_6554_pad_type_0, strides = var_6550, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_303_cast_fp16)[name = tensor("op_6554_cast_fp16")]; tensor inputs_47_cast_fp16 = add(x = var_6554_cast_fp16, y = inputs_45_cast_fp16)[name = tensor("inputs_47_cast_fp16")]; tensor var_6558 = const()[name = tensor("op_6558"), val = tensor([1])]; tensor channels_mean_47_cast_fp16 = reduce_mean(axes = var_6558, keep_dims = var_5594, x = inputs_47_cast_fp16)[name = tensor("channels_mean_47_cast_fp16")]; tensor zero_mean_47_cast_fp16 = sub(x = inputs_47_cast_fp16, y = channels_mean_47_cast_fp16)[name = tensor("zero_mean_47_cast_fp16")]; tensor zero_mean_sq_47_cast_fp16 = mul(x = zero_mean_47_cast_fp16, y = zero_mean_47_cast_fp16)[name = tensor("zero_mean_sq_47_cast_fp16")]; tensor var_6562 = const()[name = tensor("op_6562"), val = tensor([1])]; tensor var_6563_cast_fp16 = reduce_mean(axes = var_6562, keep_dims = var_5594, x = zero_mean_sq_47_cast_fp16)[name = tensor("op_6563_cast_fp16")]; tensor var_6564_to_fp16 = const()[name = tensor("op_6564_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_6565_cast_fp16 = add(x = var_6563_cast_fp16, y = var_6564_to_fp16)[name = tensor("op_6565_cast_fp16")]; tensor denom_47_epsilon_0_to_fp16 = const()[name = tensor("denom_47_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_47_cast_fp16 = rsqrt(epsilon = denom_47_epsilon_0_to_fp16, x = var_6565_cast_fp16)[name = tensor("denom_47_cast_fp16")]; tensor out_47_cast_fp16 = mul(x = zero_mean_47_cast_fp16, y = denom_47_cast_fp16)[name = tensor("out_47_cast_fp16")]; tensor var_6569_to_fp16 = const()[name = tensor("op_6569_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260935360)))]; tensor var_6570_cast_fp16 = add(x = out_47_cast_fp16, y = var_6569_to_fp16)[name = tensor("op_6570_cast_fp16")]; tensor var_6572_to_fp16 = const()[name = tensor("op_6572_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260937984)))]; tensor input_305_cast_fp16 = mul(x = var_6570_cast_fp16, y = var_6572_to_fp16)[name = tensor("input_305_cast_fp16")]; tensor var_6580 = const()[name = tensor("op_6580"), val = tensor([1, 1])]; tensor var_6582 = const()[name = tensor("op_6582"), val = tensor([1, 1])]; tensor var_6584_pad_type_0 = const()[name = tensor("op_6584_pad_type_0"), val = tensor("custom")]; tensor var_6584_pad_0 = const()[name = tensor("op_6584_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260940608)))]; tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(287155072)))]; tensor var_6584_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_6582, groups = var_5599, pad = var_6584_pad_0, pad_type = var_6584_pad_type_0, strides = var_6580, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_305_cast_fp16)[name = tensor("op_6584_cast_fp16")]; tensor var_6585_split_sizes_0 = const()[name = tensor("op_6585_split_sizes_0"), val = tensor([5120, 5120])]; tensor var_6585_axis_0 = const()[name = tensor("op_6585_axis_0"), val = tensor(1)]; tensor var_6585_cast_fp16_0, tensor var_6585_cast_fp16_1 = split(axis = var_6585_axis_0, split_sizes = var_6585_split_sizes_0, x = var_6584_cast_fp16)[name = tensor("op_6585_cast_fp16")]; tensor var_6587_mode_0 = const()[name = tensor("op_6587_mode_0"), val = tensor("EXACT")]; tensor var_6587_cast_fp16 = gelu(mode = var_6587_mode_0, x = var_6585_cast_fp16_1)[name = tensor("op_6587_cast_fp16")]; tensor input_307_cast_fp16 = mul(x = var_6585_cast_fp16_0, y = var_6587_cast_fp16)[name = tensor("input_307_cast_fp16")]; tensor var_6591 = const()[name = tensor("op_6591"), val = tensor([1, 1])]; tensor var_6593 = const()[name = tensor("op_6593"), val = tensor([1, 1])]; tensor var_6595_pad_type_0 = const()[name = tensor("op_6595_pad_type_0"), val = tensor("custom")]; tensor var_6595_pad_0 = const()[name = tensor("op_6595_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(287175616)))]; tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300282880)))]; tensor var_6595_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_6593, groups = var_5599, pad = var_6595_pad_0, pad_type = var_6595_pad_type_0, strides = var_6591, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_307_cast_fp16)[name = tensor("op_6595_cast_fp16")]; tensor hidden_states_173_cast_fp16 = add(x = var_6595_cast_fp16, y = inputs_47_cast_fp16)[name = tensor("hidden_states_173_cast_fp16")]; tensor var_6597 = const()[name = tensor("op_6597"), val = tensor([2, 1280, 12, 12])]; tensor input_309_cast_fp16 = reshape(shape = var_6597, x = hidden_states_173_cast_fp16)[name = tensor("input_309_cast_fp16")]; tensor var_6601 = const()[name = tensor("op_6601"), val = tensor([1, 1])]; tensor var_6603 = const()[name = tensor("op_6603"), val = tensor([1, 1])]; tensor hidden_states_175_pad_type_0 = const()[name = tensor("hidden_states_175_pad_type_0"), val = tensor("custom")]; tensor hidden_states_175_pad_0 = const()[name = tensor("hidden_states_175_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300285504)))]; tensor up_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303562368)))]; tensor hidden_states_175_cast_fp16 = conv(bias = up_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_6603, groups = var_5599, pad = hidden_states_175_pad_0, pad_type = hidden_states_175_pad_type_0, strides = var_6601, weight = up_blocks_1_attentions_0_proj_out_weight_to_fp16, x = input_309_cast_fp16)[name = tensor("hidden_states_175_cast_fp16")]; tensor hidden_states_177_cast_fp16 = add(x = hidden_states_175_cast_fp16, y = hidden_states_163_cast_fp16)[name = tensor("hidden_states_177_cast_fp16")]; tensor input_311_interleave_0 = const()[name = tensor("input_311_interleave_0"), val = tensor(false)]; tensor cast_5 = cast(dtype = cast_3_dtype_0, x = input_143_cast_fp16)[name = tensor("cast_5")]; tensor input_311_cast_fp16 = concat(axis = var_5599, interleave = input_311_interleave_0, values = (hidden_states_177_cast_fp16, cast_5))[name = tensor("input_311_cast_fp16")]; tensor reshape_144_shape_0 = const()[name = tensor("reshape_144_shape_0"), val = tensor([2, 32, 80, 12, 12])]; tensor reshape_144_cast_fp16 = reshape(shape = reshape_144_shape_0, x = input_311_cast_fp16)[name = tensor("reshape_144_cast_fp16")]; tensor reduce_mean_108_axes_0 = const()[name = tensor("reduce_mean_108_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_108_keep_dims_0 = const()[name = tensor("reduce_mean_108_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_108_cast_fp16 = reduce_mean(axes = reduce_mean_108_axes_0, keep_dims = reduce_mean_108_keep_dims_0, x = reshape_144_cast_fp16)[name = tensor("reduce_mean_108_cast_fp16")]; tensor sub_72_cast_fp16 = sub(x = reshape_144_cast_fp16, y = reduce_mean_108_cast_fp16)[name = tensor("sub_72_cast_fp16")]; tensor square_36_cast_fp16 = square(x = sub_72_cast_fp16)[name = tensor("square_36_cast_fp16")]; tensor reduce_mean_110_axes_0 = const()[name = tensor("reduce_mean_110_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_110_keep_dims_0 = const()[name = tensor("reduce_mean_110_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_110_cast_fp16 = reduce_mean(axes = reduce_mean_110_axes_0, keep_dims = reduce_mean_110_keep_dims_0, x = square_36_cast_fp16)[name = tensor("reduce_mean_110_cast_fp16")]; tensor add_72_y_0_to_fp16 = const()[name = tensor("add_72_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_72_cast_fp16 = add(x = reduce_mean_110_cast_fp16, y = add_72_y_0_to_fp16)[name = tensor("add_72_cast_fp16")]; tensor sqrt_36_cast_fp16 = sqrt(x = add_72_cast_fp16)[name = tensor("sqrt_36_cast_fp16")]; tensor real_div_36_cast_fp16 = real_div(x = sub_72_cast_fp16, y = sqrt_36_cast_fp16)[name = tensor("real_div_36_cast_fp16")]; tensor reshape_145_shape_0 = const()[name = tensor("reshape_145_shape_0"), val = tensor([2, 2560, 12, 12])]; tensor reshape_145_cast_fp16 = reshape(shape = reshape_145_shape_0, x = real_div_36_cast_fp16)[name = tensor("reshape_145_cast_fp16")]; tensor add_73_gamma_0_to_fp16 = const()[name = tensor("add_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303564992)))]; tensor add_73_beta_0_to_fp16 = const()[name = tensor("add_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303570176)))]; tensor add_73_epsilon_0_to_fp16 = const()[name = tensor("add_73_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_73_cast_fp16 = batch_norm(beta = add_73_beta_0_to_fp16, epsilon = add_73_epsilon_0_to_fp16, gamma = add_73_gamma_0_to_fp16, mean = add_55_mean_0_to_fp16, variance = add_55_variance_0_to_fp16, x = reshape_145_cast_fp16)[name = tensor("add_73_cast_fp16")]; tensor input_315_cast_fp16 = silu(x = add_73_cast_fp16)[name = tensor("input_315_cast_fp16")]; tensor var_6621 = const()[name = tensor("op_6621"), val = tensor([1, 1])]; tensor var_6623 = const()[name = tensor("op_6623"), val = tensor([1, 1])]; tensor hidden_states_179_pad_type_0 = const()[name = tensor("hidden_states_179_pad_type_0"), val = tensor("custom")]; tensor hidden_states_179_pad_0 = const()[name = tensor("hidden_states_179_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_1_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303575360)))]; tensor up_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362557824)))]; tensor hidden_states_179_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_6623, groups = var_5599, pad = hidden_states_179_pad_0, pad_type = hidden_states_179_pad_type_0, strides = var_6621, weight = up_blocks_1_resnets_1_conv1_weight_to_fp16, x = input_315_cast_fp16)[name = tensor("hidden_states_179_cast_fp16")]; tensor var_6629 = const()[name = tensor("op_6629"), val = tensor([1, 1])]; tensor var_6631 = const()[name = tensor("op_6631"), val = tensor([1, 1])]; tensor temb_29_pad_type_0 = const()[name = tensor("temb_29_pad_type_0"), val = tensor("custom")]; tensor temb_29_pad_0 = const()[name = tensor("temb_29_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362560448)))]; tensor up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365837312)))]; tensor temb_29_cast_fp16 = conv(bias = up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_6631, groups = var_5599, pad = temb_29_pad_0, pad_type = temb_29_pad_type_0, strides = var_6629, weight = up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16, x = cast_12)[name = tensor("temb_29_cast_fp16")]; tensor input_319_cast_fp16 = add(x = hidden_states_179_cast_fp16, y = temb_29_cast_fp16)[name = tensor("input_319_cast_fp16")]; tensor reshape_148_shape_0 = const()[name = tensor("reshape_148_shape_0"), val = tensor([2, 32, 40, 12, 12])]; tensor reshape_148_cast_fp16 = reshape(shape = reshape_148_shape_0, x = input_319_cast_fp16)[name = tensor("reshape_148_cast_fp16")]; tensor reduce_mean_111_axes_0 = const()[name = tensor("reduce_mean_111_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_111_keep_dims_0 = const()[name = tensor("reduce_mean_111_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_111_cast_fp16 = reduce_mean(axes = reduce_mean_111_axes_0, keep_dims = reduce_mean_111_keep_dims_0, x = reshape_148_cast_fp16)[name = tensor("reduce_mean_111_cast_fp16")]; tensor sub_74_cast_fp16 = sub(x = reshape_148_cast_fp16, y = reduce_mean_111_cast_fp16)[name = tensor("sub_74_cast_fp16")]; tensor square_37_cast_fp16 = square(x = sub_74_cast_fp16)[name = tensor("square_37_cast_fp16")]; tensor reduce_mean_113_axes_0 = const()[name = tensor("reduce_mean_113_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_113_keep_dims_0 = const()[name = tensor("reduce_mean_113_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_113_cast_fp16 = reduce_mean(axes = reduce_mean_113_axes_0, keep_dims = reduce_mean_113_keep_dims_0, x = square_37_cast_fp16)[name = tensor("reduce_mean_113_cast_fp16")]; tensor add_74_y_0_to_fp16 = const()[name = tensor("add_74_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_74_cast_fp16 = add(x = reduce_mean_113_cast_fp16, y = add_74_y_0_to_fp16)[name = tensor("add_74_cast_fp16")]; tensor sqrt_37_cast_fp16 = sqrt(x = add_74_cast_fp16)[name = tensor("sqrt_37_cast_fp16")]; tensor real_div_37_cast_fp16 = real_div(x = sub_74_cast_fp16, y = sqrt_37_cast_fp16)[name = tensor("real_div_37_cast_fp16")]; tensor reshape_149_shape_0 = const()[name = tensor("reshape_149_shape_0"), val = tensor([2, 1280, 12, 12])]; tensor reshape_149_cast_fp16 = reshape(shape = reshape_149_shape_0, x = real_div_37_cast_fp16)[name = tensor("reshape_149_cast_fp16")]; tensor add_75_gamma_0_to_fp16 = const()[name = tensor("add_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365839936)))]; tensor add_75_beta_0_to_fp16 = const()[name = tensor("add_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365842560)))]; tensor add_75_epsilon_0_to_fp16 = const()[name = tensor("add_75_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_75_cast_fp16 = batch_norm(beta = add_75_beta_0_to_fp16, epsilon = add_75_epsilon_0_to_fp16, gamma = add_75_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_149_cast_fp16)[name = tensor("add_75_cast_fp16")]; tensor input_323_cast_fp16 = silu(x = add_75_cast_fp16)[name = tensor("input_323_cast_fp16")]; tensor var_6641 = const()[name = tensor("op_6641"), val = tensor([1, 1])]; tensor var_6643 = const()[name = tensor("op_6643"), val = tensor([1, 1])]; tensor hidden_states_181_pad_type_0 = const()[name = tensor("hidden_states_181_pad_type_0"), val = tensor("custom")]; tensor hidden_states_181_pad_0 = const()[name = tensor("hidden_states_181_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_1_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365845184)))]; tensor up_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395336448)))]; tensor hidden_states_181_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_6643, groups = var_5599, pad = hidden_states_181_pad_0, pad_type = hidden_states_181_pad_type_0, strides = var_6641, weight = up_blocks_1_resnets_1_conv2_weight_to_fp16, x = input_323_cast_fp16)[name = tensor("hidden_states_181_cast_fp16")]; tensor var_6648 = const()[name = tensor("op_6648"), val = tensor([1, 1])]; tensor var_6650 = const()[name = tensor("op_6650"), val = tensor([1, 1])]; tensor x_13_pad_type_0 = const()[name = tensor("x_13_pad_type_0"), val = tensor("custom")]; tensor x_13_pad_0 = const()[name = tensor("x_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395339072)))]; tensor up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401892736)))]; tensor x_13_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_6650, groups = var_5599, pad = x_13_pad_0, pad_type = x_13_pad_type_0, strides = var_6648, weight = up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16, x = input_311_cast_fp16)[name = tensor("x_13_cast_fp16")]; tensor hidden_states_183_cast_fp16 = add(x = x_13_cast_fp16, y = hidden_states_181_cast_fp16)[name = tensor("hidden_states_183_cast_fp16")]; tensor reshape_152_shape_0 = const()[name = tensor("reshape_152_shape_0"), val = tensor([2, 32, 40, 12, 12])]; tensor reshape_152_cast_fp16 = reshape(shape = reshape_152_shape_0, x = hidden_states_183_cast_fp16)[name = tensor("reshape_152_cast_fp16")]; tensor reduce_mean_114_axes_0 = const()[name = tensor("reduce_mean_114_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_114_keep_dims_0 = const()[name = tensor("reduce_mean_114_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_114_cast_fp16 = reduce_mean(axes = reduce_mean_114_axes_0, keep_dims = reduce_mean_114_keep_dims_0, x = reshape_152_cast_fp16)[name = tensor("reduce_mean_114_cast_fp16")]; tensor sub_76_cast_fp16 = sub(x = reshape_152_cast_fp16, y = reduce_mean_114_cast_fp16)[name = tensor("sub_76_cast_fp16")]; tensor square_38_cast_fp16 = square(x = sub_76_cast_fp16)[name = tensor("square_38_cast_fp16")]; tensor reduce_mean_116_axes_0 = const()[name = tensor("reduce_mean_116_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_116_keep_dims_0 = const()[name = tensor("reduce_mean_116_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_116_cast_fp16 = reduce_mean(axes = reduce_mean_116_axes_0, keep_dims = reduce_mean_116_keep_dims_0, x = square_38_cast_fp16)[name = tensor("reduce_mean_116_cast_fp16")]; tensor add_76_y_0_to_fp16 = const()[name = tensor("add_76_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_76_cast_fp16 = add(x = reduce_mean_116_cast_fp16, y = add_76_y_0_to_fp16)[name = tensor("add_76_cast_fp16")]; tensor sqrt_38_cast_fp16 = sqrt(x = add_76_cast_fp16)[name = tensor("sqrt_38_cast_fp16")]; tensor real_div_38_cast_fp16 = real_div(x = sub_76_cast_fp16, y = sqrt_38_cast_fp16)[name = tensor("real_div_38_cast_fp16")]; tensor reshape_153_shape_0 = const()[name = tensor("reshape_153_shape_0"), val = tensor([2, 1280, 12, 12])]; tensor reshape_153_cast_fp16 = reshape(shape = reshape_153_shape_0, x = real_div_38_cast_fp16)[name = tensor("reshape_153_cast_fp16")]; tensor add_77_gamma_0_to_fp16 = const()[name = tensor("add_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401895360)))]; tensor add_77_beta_0_to_fp16 = const()[name = tensor("add_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401897984)))]; tensor add_77_epsilon_0_to_fp16 = const()[name = tensor("add_77_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_77_cast_fp16 = batch_norm(beta = add_77_beta_0_to_fp16, epsilon = add_77_epsilon_0_to_fp16, gamma = add_77_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_153_cast_fp16)[name = tensor("add_77_cast_fp16")]; tensor var_6670 = const()[name = tensor("op_6670"), val = tensor([1, 1])]; tensor var_6672 = const()[name = tensor("op_6672"), val = tensor([1, 1])]; tensor hidden_states_185_pad_type_0 = const()[name = tensor("hidden_states_185_pad_type_0"), val = tensor("custom")]; tensor hidden_states_185_pad_0 = const()[name = tensor("hidden_states_185_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(401900608)))]; tensor up_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405177472)))]; tensor hidden_states_185_cast_fp16 = conv(bias = up_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_6672, groups = var_5599, pad = hidden_states_185_pad_0, pad_type = hidden_states_185_pad_type_0, strides = var_6670, weight = up_blocks_1_attentions_1_proj_in_weight_to_fp16, x = add_77_cast_fp16)[name = tensor("hidden_states_185_cast_fp16")]; tensor var_6677 = const()[name = tensor("op_6677"), val = tensor([2, 1280, 1, 144])]; tensor inputs_49_cast_fp16 = reshape(shape = var_6677, x = hidden_states_185_cast_fp16)[name = tensor("inputs_49_cast_fp16")]; tensor var_6687 = const()[name = tensor("op_6687"), val = tensor([1])]; tensor channels_mean_49_cast_fp16 = reduce_mean(axes = var_6687, keep_dims = var_5594, x = inputs_49_cast_fp16)[name = tensor("channels_mean_49_cast_fp16")]; tensor zero_mean_49_cast_fp16 = sub(x = inputs_49_cast_fp16, y = channels_mean_49_cast_fp16)[name = tensor("zero_mean_49_cast_fp16")]; tensor zero_mean_sq_49_cast_fp16 = mul(x = zero_mean_49_cast_fp16, y = zero_mean_49_cast_fp16)[name = tensor("zero_mean_sq_49_cast_fp16")]; tensor var_6691 = const()[name = tensor("op_6691"), val = tensor([1])]; tensor var_6692_cast_fp16 = reduce_mean(axes = var_6691, keep_dims = var_5594, x = zero_mean_sq_49_cast_fp16)[name = tensor("op_6692_cast_fp16")]; tensor var_6693_to_fp16 = const()[name = tensor("op_6693_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_6694_cast_fp16 = add(x = var_6692_cast_fp16, y = var_6693_to_fp16)[name = tensor("op_6694_cast_fp16")]; tensor denom_49_epsilon_0_to_fp16 = const()[name = tensor("denom_49_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_49_cast_fp16 = rsqrt(epsilon = denom_49_epsilon_0_to_fp16, x = var_6694_cast_fp16)[name = tensor("denom_49_cast_fp16")]; tensor out_49_cast_fp16 = mul(x = zero_mean_49_cast_fp16, y = denom_49_cast_fp16)[name = tensor("out_49_cast_fp16")]; tensor var_6698_to_fp16 = const()[name = tensor("op_6698_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405180096)))]; tensor var_6699_cast_fp16 = add(x = out_49_cast_fp16, y = var_6698_to_fp16)[name = tensor("op_6699_cast_fp16")]; tensor var_6701_to_fp16 = const()[name = tensor("op_6701_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405182720)))]; tensor hidden_states_187_cast_fp16 = mul(x = var_6699_cast_fp16, y = var_6701_to_fp16)[name = tensor("hidden_states_187_cast_fp16")]; tensor var_6708 = const()[name = tensor("op_6708"), val = tensor([1, 1])]; tensor var_6710 = const()[name = tensor("op_6710"), val = tensor([1, 1])]; tensor q_33_pad_type_0 = const()[name = tensor("q_33_pad_type_0"), val = tensor("custom")]; tensor q_33_pad_0 = const()[name = tensor("q_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405185344)))]; tensor q_33_cast_fp16 = conv(dilations = var_6710, groups = var_5599, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = var_6708, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_187_cast_fp16)[name = tensor("q_33_cast_fp16")]; tensor var_6714 = const()[name = tensor("op_6714"), val = tensor([1, 1])]; tensor var_6716 = const()[name = tensor("op_6716"), val = tensor([1, 1])]; tensor k_65_pad_type_0 = const()[name = tensor("k_65_pad_type_0"), val = tensor("custom")]; tensor k_65_pad_0 = const()[name = tensor("k_65_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408462208)))]; tensor k_65_cast_fp16 = conv(dilations = var_6716, groups = var_5599, pad = k_65_pad_0, pad_type = k_65_pad_type_0, strides = var_6714, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_187_cast_fp16)[name = tensor("k_65_cast_fp16")]; tensor var_6720 = const()[name = tensor("op_6720"), val = tensor([1, 1])]; tensor var_6722 = const()[name = tensor("op_6722"), val = tensor([1, 1])]; tensor v_33_pad_type_0 = const()[name = tensor("v_33_pad_type_0"), val = tensor("custom")]; tensor v_33_pad_0 = const()[name = tensor("v_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411739072)))]; tensor v_33_cast_fp16 = conv(dilations = var_6722, groups = var_5599, pad = v_33_pad_0, pad_type = v_33_pad_type_0, strides = var_6720, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_187_cast_fp16)[name = tensor("v_33_cast_fp16")]; tensor var_6726_begin_0 = const()[name = tensor("op_6726_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6726_end_0 = const()[name = tensor("op_6726_end_0"), val = tensor([2, 64, 1, 144])]; tensor var_6726_end_mask_0 = const()[name = tensor("op_6726_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6726_cast_fp16 = slice_by_index(begin = var_6726_begin_0, end = var_6726_end_0, end_mask = var_6726_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6726_cast_fp16")]; tensor var_6730_begin_0 = const()[name = tensor("op_6730_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_6730_end_0 = const()[name = tensor("op_6730_end_0"), val = tensor([2, 128, 1, 144])]; tensor var_6730_end_mask_0 = const()[name = tensor("op_6730_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6730_cast_fp16 = slice_by_index(begin = var_6730_begin_0, end = var_6730_end_0, end_mask = var_6730_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6730_cast_fp16")]; tensor var_6734_begin_0 = const()[name = tensor("op_6734_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_6734_end_0 = const()[name = tensor("op_6734_end_0"), val = tensor([2, 192, 1, 144])]; tensor var_6734_end_mask_0 = const()[name = tensor("op_6734_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6734_cast_fp16 = slice_by_index(begin = var_6734_begin_0, end = var_6734_end_0, end_mask = var_6734_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6734_cast_fp16")]; tensor var_6738_begin_0 = const()[name = tensor("op_6738_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_6738_end_0 = const()[name = tensor("op_6738_end_0"), val = tensor([2, 256, 1, 144])]; tensor var_6738_end_mask_0 = const()[name = tensor("op_6738_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6738_cast_fp16 = slice_by_index(begin = var_6738_begin_0, end = var_6738_end_0, end_mask = var_6738_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6738_cast_fp16")]; tensor var_6742_begin_0 = const()[name = tensor("op_6742_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_6742_end_0 = const()[name = tensor("op_6742_end_0"), val = tensor([2, 320, 1, 144])]; tensor var_6742_end_mask_0 = const()[name = tensor("op_6742_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6742_cast_fp16 = slice_by_index(begin = var_6742_begin_0, end = var_6742_end_0, end_mask = var_6742_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6742_cast_fp16")]; tensor var_6746_begin_0 = const()[name = tensor("op_6746_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6746_end_0 = const()[name = tensor("op_6746_end_0"), val = tensor([2, 384, 1, 144])]; tensor var_6746_end_mask_0 = const()[name = tensor("op_6746_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6746_cast_fp16 = slice_by_index(begin = var_6746_begin_0, end = var_6746_end_0, end_mask = var_6746_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6746_cast_fp16")]; tensor var_6750_begin_0 = const()[name = tensor("op_6750_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_6750_end_0 = const()[name = tensor("op_6750_end_0"), val = tensor([2, 448, 1, 144])]; tensor var_6750_end_mask_0 = const()[name = tensor("op_6750_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6750_cast_fp16 = slice_by_index(begin = var_6750_begin_0, end = var_6750_end_0, end_mask = var_6750_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6750_cast_fp16")]; tensor var_6754_begin_0 = const()[name = tensor("op_6754_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_6754_end_0 = const()[name = tensor("op_6754_end_0"), val = tensor([2, 512, 1, 144])]; tensor var_6754_end_mask_0 = const()[name = tensor("op_6754_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6754_cast_fp16 = slice_by_index(begin = var_6754_begin_0, end = var_6754_end_0, end_mask = var_6754_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6754_cast_fp16")]; tensor var_6758_begin_0 = const()[name = tensor("op_6758_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_6758_end_0 = const()[name = tensor("op_6758_end_0"), val = tensor([2, 576, 1, 144])]; tensor var_6758_end_mask_0 = const()[name = tensor("op_6758_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6758_cast_fp16 = slice_by_index(begin = var_6758_begin_0, end = var_6758_end_0, end_mask = var_6758_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6758_cast_fp16")]; tensor var_6762_begin_0 = const()[name = tensor("op_6762_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_6762_end_0 = const()[name = tensor("op_6762_end_0"), val = tensor([2, 640, 1, 144])]; tensor var_6762_end_mask_0 = const()[name = tensor("op_6762_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6762_cast_fp16 = slice_by_index(begin = var_6762_begin_0, end = var_6762_end_0, end_mask = var_6762_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6762_cast_fp16")]; tensor var_6766_begin_0 = const()[name = tensor("op_6766_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6766_end_0 = const()[name = tensor("op_6766_end_0"), val = tensor([2, 704, 1, 144])]; tensor var_6766_end_mask_0 = const()[name = tensor("op_6766_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6766_cast_fp16 = slice_by_index(begin = var_6766_begin_0, end = var_6766_end_0, end_mask = var_6766_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6766_cast_fp16")]; tensor var_6770_begin_0 = const()[name = tensor("op_6770_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_6770_end_0 = const()[name = tensor("op_6770_end_0"), val = tensor([2, 768, 1, 144])]; tensor var_6770_end_mask_0 = const()[name = tensor("op_6770_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6770_cast_fp16 = slice_by_index(begin = var_6770_begin_0, end = var_6770_end_0, end_mask = var_6770_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6770_cast_fp16")]; tensor var_6774_begin_0 = const()[name = tensor("op_6774_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_6774_end_0 = const()[name = tensor("op_6774_end_0"), val = tensor([2, 832, 1, 144])]; tensor var_6774_end_mask_0 = const()[name = tensor("op_6774_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6774_cast_fp16 = slice_by_index(begin = var_6774_begin_0, end = var_6774_end_0, end_mask = var_6774_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6774_cast_fp16")]; tensor var_6778_begin_0 = const()[name = tensor("op_6778_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_6778_end_0 = const()[name = tensor("op_6778_end_0"), val = tensor([2, 896, 1, 144])]; tensor var_6778_end_mask_0 = const()[name = tensor("op_6778_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6778_cast_fp16 = slice_by_index(begin = var_6778_begin_0, end = var_6778_end_0, end_mask = var_6778_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6778_cast_fp16")]; tensor var_6782_begin_0 = const()[name = tensor("op_6782_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_6782_end_0 = const()[name = tensor("op_6782_end_0"), val = tensor([2, 960, 1, 144])]; tensor var_6782_end_mask_0 = const()[name = tensor("op_6782_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6782_cast_fp16 = slice_by_index(begin = var_6782_begin_0, end = var_6782_end_0, end_mask = var_6782_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6782_cast_fp16")]; tensor var_6786_begin_0 = const()[name = tensor("op_6786_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_6786_end_0 = const()[name = tensor("op_6786_end_0"), val = tensor([2, 1024, 1, 144])]; tensor var_6786_end_mask_0 = const()[name = tensor("op_6786_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6786_cast_fp16 = slice_by_index(begin = var_6786_begin_0, end = var_6786_end_0, end_mask = var_6786_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6786_cast_fp16")]; tensor var_6790_begin_0 = const()[name = tensor("op_6790_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_6790_end_0 = const()[name = tensor("op_6790_end_0"), val = tensor([2, 1088, 1, 144])]; tensor var_6790_end_mask_0 = const()[name = tensor("op_6790_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6790_cast_fp16 = slice_by_index(begin = var_6790_begin_0, end = var_6790_end_0, end_mask = var_6790_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6790_cast_fp16")]; tensor var_6794_begin_0 = const()[name = tensor("op_6794_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_6794_end_0 = const()[name = tensor("op_6794_end_0"), val = tensor([2, 1152, 1, 144])]; tensor var_6794_end_mask_0 = const()[name = tensor("op_6794_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6794_cast_fp16 = slice_by_index(begin = var_6794_begin_0, end = var_6794_end_0, end_mask = var_6794_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6794_cast_fp16")]; tensor var_6798_begin_0 = const()[name = tensor("op_6798_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_6798_end_0 = const()[name = tensor("op_6798_end_0"), val = tensor([2, 1216, 1, 144])]; tensor var_6798_end_mask_0 = const()[name = tensor("op_6798_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6798_cast_fp16 = slice_by_index(begin = var_6798_begin_0, end = var_6798_end_0, end_mask = var_6798_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6798_cast_fp16")]; tensor var_6802_begin_0 = const()[name = tensor("op_6802_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_6802_end_0 = const()[name = tensor("op_6802_end_0"), val = tensor([2, 1280, 1, 144])]; tensor var_6802_end_mask_0 = const()[name = tensor("op_6802_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6802_cast_fp16 = slice_by_index(begin = var_6802_begin_0, end = var_6802_end_0, end_mask = var_6802_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_6802_cast_fp16")]; tensor k_67_perm_0 = const()[name = tensor("k_67_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_6809_begin_0 = const()[name = tensor("op_6809_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6809_end_0 = const()[name = tensor("op_6809_end_0"), val = tensor([2, 144, 1, 64])]; tensor var_6809_end_mask_0 = const()[name = tensor("op_6809_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_15 = transpose(perm = k_67_perm_0, x = k_65_cast_fp16)[name = tensor("transpose_15")]; tensor var_6809_cast_fp16 = slice_by_index(begin = var_6809_begin_0, end = var_6809_end_0, end_mask = var_6809_end_mask_0, x = transpose_15)[name = tensor("op_6809_cast_fp16")]; tensor var_6813_begin_0 = const()[name = tensor("op_6813_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_6813_end_0 = const()[name = tensor("op_6813_end_0"), val = tensor([2, 144, 1, 128])]; tensor var_6813_end_mask_0 = const()[name = tensor("op_6813_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6813_cast_fp16 = slice_by_index(begin = var_6813_begin_0, end = var_6813_end_0, end_mask = var_6813_end_mask_0, x = transpose_15)[name = tensor("op_6813_cast_fp16")]; tensor var_6817_begin_0 = const()[name = tensor("op_6817_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_6817_end_0 = const()[name = tensor("op_6817_end_0"), val = tensor([2, 144, 1, 192])]; tensor var_6817_end_mask_0 = const()[name = tensor("op_6817_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6817_cast_fp16 = slice_by_index(begin = var_6817_begin_0, end = var_6817_end_0, end_mask = var_6817_end_mask_0, x = transpose_15)[name = tensor("op_6817_cast_fp16")]; tensor var_6821_begin_0 = const()[name = tensor("op_6821_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_6821_end_0 = const()[name = tensor("op_6821_end_0"), val = tensor([2, 144, 1, 256])]; tensor var_6821_end_mask_0 = const()[name = tensor("op_6821_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6821_cast_fp16 = slice_by_index(begin = var_6821_begin_0, end = var_6821_end_0, end_mask = var_6821_end_mask_0, x = transpose_15)[name = tensor("op_6821_cast_fp16")]; tensor var_6825_begin_0 = const()[name = tensor("op_6825_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_6825_end_0 = const()[name = tensor("op_6825_end_0"), val = tensor([2, 144, 1, 320])]; tensor var_6825_end_mask_0 = const()[name = tensor("op_6825_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6825_cast_fp16 = slice_by_index(begin = var_6825_begin_0, end = var_6825_end_0, end_mask = var_6825_end_mask_0, x = transpose_15)[name = tensor("op_6825_cast_fp16")]; tensor var_6829_begin_0 = const()[name = tensor("op_6829_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_6829_end_0 = const()[name = tensor("op_6829_end_0"), val = tensor([2, 144, 1, 384])]; tensor var_6829_end_mask_0 = const()[name = tensor("op_6829_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6829_cast_fp16 = slice_by_index(begin = var_6829_begin_0, end = var_6829_end_0, end_mask = var_6829_end_mask_0, x = transpose_15)[name = tensor("op_6829_cast_fp16")]; tensor var_6833_begin_0 = const()[name = tensor("op_6833_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_6833_end_0 = const()[name = tensor("op_6833_end_0"), val = tensor([2, 144, 1, 448])]; tensor var_6833_end_mask_0 = const()[name = tensor("op_6833_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6833_cast_fp16 = slice_by_index(begin = var_6833_begin_0, end = var_6833_end_0, end_mask = var_6833_end_mask_0, x = transpose_15)[name = tensor("op_6833_cast_fp16")]; tensor var_6837_begin_0 = const()[name = tensor("op_6837_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_6837_end_0 = const()[name = tensor("op_6837_end_0"), val = tensor([2, 144, 1, 512])]; tensor var_6837_end_mask_0 = const()[name = tensor("op_6837_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6837_cast_fp16 = slice_by_index(begin = var_6837_begin_0, end = var_6837_end_0, end_mask = var_6837_end_mask_0, x = transpose_15)[name = tensor("op_6837_cast_fp16")]; tensor var_6841_begin_0 = const()[name = tensor("op_6841_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6841_end_0 = const()[name = tensor("op_6841_end_0"), val = tensor([2, 144, 1, 576])]; tensor var_6841_end_mask_0 = const()[name = tensor("op_6841_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6841_cast_fp16 = slice_by_index(begin = var_6841_begin_0, end = var_6841_end_0, end_mask = var_6841_end_mask_0, x = transpose_15)[name = tensor("op_6841_cast_fp16")]; tensor var_6845_begin_0 = const()[name = tensor("op_6845_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_6845_end_0 = const()[name = tensor("op_6845_end_0"), val = tensor([2, 144, 1, 640])]; tensor var_6845_end_mask_0 = const()[name = tensor("op_6845_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6845_cast_fp16 = slice_by_index(begin = var_6845_begin_0, end = var_6845_end_0, end_mask = var_6845_end_mask_0, x = transpose_15)[name = tensor("op_6845_cast_fp16")]; tensor var_6849_begin_0 = const()[name = tensor("op_6849_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_6849_end_0 = const()[name = tensor("op_6849_end_0"), val = tensor([2, 144, 1, 704])]; tensor var_6849_end_mask_0 = const()[name = tensor("op_6849_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6849_cast_fp16 = slice_by_index(begin = var_6849_begin_0, end = var_6849_end_0, end_mask = var_6849_end_mask_0, x = transpose_15)[name = tensor("op_6849_cast_fp16")]; tensor var_6853_begin_0 = const()[name = tensor("op_6853_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_6853_end_0 = const()[name = tensor("op_6853_end_0"), val = tensor([2, 144, 1, 768])]; tensor var_6853_end_mask_0 = const()[name = tensor("op_6853_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6853_cast_fp16 = slice_by_index(begin = var_6853_begin_0, end = var_6853_end_0, end_mask = var_6853_end_mask_0, x = transpose_15)[name = tensor("op_6853_cast_fp16")]; tensor var_6857_begin_0 = const()[name = tensor("op_6857_begin_0"), val = tensor([0, 0, 0, 768])]; tensor var_6857_end_0 = const()[name = tensor("op_6857_end_0"), val = tensor([2, 144, 1, 832])]; tensor var_6857_end_mask_0 = const()[name = tensor("op_6857_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6857_cast_fp16 = slice_by_index(begin = var_6857_begin_0, end = var_6857_end_0, end_mask = var_6857_end_mask_0, x = transpose_15)[name = tensor("op_6857_cast_fp16")]; tensor var_6861_begin_0 = const()[name = tensor("op_6861_begin_0"), val = tensor([0, 0, 0, 832])]; tensor var_6861_end_0 = const()[name = tensor("op_6861_end_0"), val = tensor([2, 144, 1, 896])]; tensor var_6861_end_mask_0 = const()[name = tensor("op_6861_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6861_cast_fp16 = slice_by_index(begin = var_6861_begin_0, end = var_6861_end_0, end_mask = var_6861_end_mask_0, x = transpose_15)[name = tensor("op_6861_cast_fp16")]; tensor var_6865_begin_0 = const()[name = tensor("op_6865_begin_0"), val = tensor([0, 0, 0, 896])]; tensor var_6865_end_0 = const()[name = tensor("op_6865_end_0"), val = tensor([2, 144, 1, 960])]; tensor var_6865_end_mask_0 = const()[name = tensor("op_6865_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6865_cast_fp16 = slice_by_index(begin = var_6865_begin_0, end = var_6865_end_0, end_mask = var_6865_end_mask_0, x = transpose_15)[name = tensor("op_6865_cast_fp16")]; tensor var_6869_begin_0 = const()[name = tensor("op_6869_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_6869_end_0 = const()[name = tensor("op_6869_end_0"), val = tensor([2, 144, 1, 1024])]; tensor var_6869_end_mask_0 = const()[name = tensor("op_6869_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6869_cast_fp16 = slice_by_index(begin = var_6869_begin_0, end = var_6869_end_0, end_mask = var_6869_end_mask_0, x = transpose_15)[name = tensor("op_6869_cast_fp16")]; tensor var_6873_begin_0 = const()[name = tensor("op_6873_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_6873_end_0 = const()[name = tensor("op_6873_end_0"), val = tensor([2, 144, 1, 1088])]; tensor var_6873_end_mask_0 = const()[name = tensor("op_6873_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6873_cast_fp16 = slice_by_index(begin = var_6873_begin_0, end = var_6873_end_0, end_mask = var_6873_end_mask_0, x = transpose_15)[name = tensor("op_6873_cast_fp16")]; tensor var_6877_begin_0 = const()[name = tensor("op_6877_begin_0"), val = tensor([0, 0, 0, 1088])]; tensor var_6877_end_0 = const()[name = tensor("op_6877_end_0"), val = tensor([2, 144, 1, 1152])]; tensor var_6877_end_mask_0 = const()[name = tensor("op_6877_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6877_cast_fp16 = slice_by_index(begin = var_6877_begin_0, end = var_6877_end_0, end_mask = var_6877_end_mask_0, x = transpose_15)[name = tensor("op_6877_cast_fp16")]; tensor var_6881_begin_0 = const()[name = tensor("op_6881_begin_0"), val = tensor([0, 0, 0, 1152])]; tensor var_6881_end_0 = const()[name = tensor("op_6881_end_0"), val = tensor([2, 144, 1, 1216])]; tensor var_6881_end_mask_0 = const()[name = tensor("op_6881_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6881_cast_fp16 = slice_by_index(begin = var_6881_begin_0, end = var_6881_end_0, end_mask = var_6881_end_mask_0, x = transpose_15)[name = tensor("op_6881_cast_fp16")]; tensor var_6885_begin_0 = const()[name = tensor("op_6885_begin_0"), val = tensor([0, 0, 0, 1216])]; tensor var_6885_end_0 = const()[name = tensor("op_6885_end_0"), val = tensor([2, 144, 1, 1280])]; tensor var_6885_end_mask_0 = const()[name = tensor("op_6885_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6885_cast_fp16 = slice_by_index(begin = var_6885_begin_0, end = var_6885_end_0, end_mask = var_6885_end_mask_0, x = transpose_15)[name = tensor("op_6885_cast_fp16")]; tensor var_6887_begin_0 = const()[name = tensor("op_6887_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6887_end_0 = const()[name = tensor("op_6887_end_0"), val = tensor([2, 64, 1, 144])]; tensor var_6887_end_mask_0 = const()[name = tensor("op_6887_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6887_cast_fp16 = slice_by_index(begin = var_6887_begin_0, end = var_6887_end_0, end_mask = var_6887_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6887_cast_fp16")]; tensor var_6891_begin_0 = const()[name = tensor("op_6891_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_6891_end_0 = const()[name = tensor("op_6891_end_0"), val = tensor([2, 128, 1, 144])]; tensor var_6891_end_mask_0 = const()[name = tensor("op_6891_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6891_cast_fp16 = slice_by_index(begin = var_6891_begin_0, end = var_6891_end_0, end_mask = var_6891_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6891_cast_fp16")]; tensor var_6895_begin_0 = const()[name = tensor("op_6895_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_6895_end_0 = const()[name = tensor("op_6895_end_0"), val = tensor([2, 192, 1, 144])]; tensor var_6895_end_mask_0 = const()[name = tensor("op_6895_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6895_cast_fp16 = slice_by_index(begin = var_6895_begin_0, end = var_6895_end_0, end_mask = var_6895_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6895_cast_fp16")]; tensor var_6899_begin_0 = const()[name = tensor("op_6899_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_6899_end_0 = const()[name = tensor("op_6899_end_0"), val = tensor([2, 256, 1, 144])]; tensor var_6899_end_mask_0 = const()[name = tensor("op_6899_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6899_cast_fp16 = slice_by_index(begin = var_6899_begin_0, end = var_6899_end_0, end_mask = var_6899_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6899_cast_fp16")]; tensor var_6903_begin_0 = const()[name = tensor("op_6903_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_6903_end_0 = const()[name = tensor("op_6903_end_0"), val = tensor([2, 320, 1, 144])]; tensor var_6903_end_mask_0 = const()[name = tensor("op_6903_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6903_cast_fp16 = slice_by_index(begin = var_6903_begin_0, end = var_6903_end_0, end_mask = var_6903_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6903_cast_fp16")]; tensor var_6907_begin_0 = const()[name = tensor("op_6907_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6907_end_0 = const()[name = tensor("op_6907_end_0"), val = tensor([2, 384, 1, 144])]; tensor var_6907_end_mask_0 = const()[name = tensor("op_6907_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6907_cast_fp16 = slice_by_index(begin = var_6907_begin_0, end = var_6907_end_0, end_mask = var_6907_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6907_cast_fp16")]; tensor var_6911_begin_0 = const()[name = tensor("op_6911_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_6911_end_0 = const()[name = tensor("op_6911_end_0"), val = tensor([2, 448, 1, 144])]; tensor var_6911_end_mask_0 = const()[name = tensor("op_6911_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6911_cast_fp16 = slice_by_index(begin = var_6911_begin_0, end = var_6911_end_0, end_mask = var_6911_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6911_cast_fp16")]; tensor var_6915_begin_0 = const()[name = tensor("op_6915_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_6915_end_0 = const()[name = tensor("op_6915_end_0"), val = tensor([2, 512, 1, 144])]; tensor var_6915_end_mask_0 = const()[name = tensor("op_6915_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6915_cast_fp16 = slice_by_index(begin = var_6915_begin_0, end = var_6915_end_0, end_mask = var_6915_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6915_cast_fp16")]; tensor var_6919_begin_0 = const()[name = tensor("op_6919_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_6919_end_0 = const()[name = tensor("op_6919_end_0"), val = tensor([2, 576, 1, 144])]; tensor var_6919_end_mask_0 = const()[name = tensor("op_6919_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6919_cast_fp16 = slice_by_index(begin = var_6919_begin_0, end = var_6919_end_0, end_mask = var_6919_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6919_cast_fp16")]; tensor var_6923_begin_0 = const()[name = tensor("op_6923_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_6923_end_0 = const()[name = tensor("op_6923_end_0"), val = tensor([2, 640, 1, 144])]; tensor var_6923_end_mask_0 = const()[name = tensor("op_6923_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6923_cast_fp16 = slice_by_index(begin = var_6923_begin_0, end = var_6923_end_0, end_mask = var_6923_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6923_cast_fp16")]; tensor var_6927_begin_0 = const()[name = tensor("op_6927_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6927_end_0 = const()[name = tensor("op_6927_end_0"), val = tensor([2, 704, 1, 144])]; tensor var_6927_end_mask_0 = const()[name = tensor("op_6927_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6927_cast_fp16 = slice_by_index(begin = var_6927_begin_0, end = var_6927_end_0, end_mask = var_6927_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6927_cast_fp16")]; tensor var_6931_begin_0 = const()[name = tensor("op_6931_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_6931_end_0 = const()[name = tensor("op_6931_end_0"), val = tensor([2, 768, 1, 144])]; tensor var_6931_end_mask_0 = const()[name = tensor("op_6931_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6931_cast_fp16 = slice_by_index(begin = var_6931_begin_0, end = var_6931_end_0, end_mask = var_6931_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6931_cast_fp16")]; tensor var_6935_begin_0 = const()[name = tensor("op_6935_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_6935_end_0 = const()[name = tensor("op_6935_end_0"), val = tensor([2, 832, 1, 144])]; tensor var_6935_end_mask_0 = const()[name = tensor("op_6935_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6935_cast_fp16 = slice_by_index(begin = var_6935_begin_0, end = var_6935_end_0, end_mask = var_6935_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6935_cast_fp16")]; tensor var_6939_begin_0 = const()[name = tensor("op_6939_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_6939_end_0 = const()[name = tensor("op_6939_end_0"), val = tensor([2, 896, 1, 144])]; tensor var_6939_end_mask_0 = const()[name = tensor("op_6939_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6939_cast_fp16 = slice_by_index(begin = var_6939_begin_0, end = var_6939_end_0, end_mask = var_6939_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6939_cast_fp16")]; tensor var_6943_begin_0 = const()[name = tensor("op_6943_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_6943_end_0 = const()[name = tensor("op_6943_end_0"), val = tensor([2, 960, 1, 144])]; tensor var_6943_end_mask_0 = const()[name = tensor("op_6943_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6943_cast_fp16 = slice_by_index(begin = var_6943_begin_0, end = var_6943_end_0, end_mask = var_6943_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6943_cast_fp16")]; tensor var_6947_begin_0 = const()[name = tensor("op_6947_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_6947_end_0 = const()[name = tensor("op_6947_end_0"), val = tensor([2, 1024, 1, 144])]; tensor var_6947_end_mask_0 = const()[name = tensor("op_6947_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6947_cast_fp16 = slice_by_index(begin = var_6947_begin_0, end = var_6947_end_0, end_mask = var_6947_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6947_cast_fp16")]; tensor var_6951_begin_0 = const()[name = tensor("op_6951_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_6951_end_0 = const()[name = tensor("op_6951_end_0"), val = tensor([2, 1088, 1, 144])]; tensor var_6951_end_mask_0 = const()[name = tensor("op_6951_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6951_cast_fp16 = slice_by_index(begin = var_6951_begin_0, end = var_6951_end_0, end_mask = var_6951_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6951_cast_fp16")]; tensor var_6955_begin_0 = const()[name = tensor("op_6955_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_6955_end_0 = const()[name = tensor("op_6955_end_0"), val = tensor([2, 1152, 1, 144])]; tensor var_6955_end_mask_0 = const()[name = tensor("op_6955_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6955_cast_fp16 = slice_by_index(begin = var_6955_begin_0, end = var_6955_end_0, end_mask = var_6955_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6955_cast_fp16")]; tensor var_6959_begin_0 = const()[name = tensor("op_6959_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_6959_end_0 = const()[name = tensor("op_6959_end_0"), val = tensor([2, 1216, 1, 144])]; tensor var_6959_end_mask_0 = const()[name = tensor("op_6959_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6959_cast_fp16 = slice_by_index(begin = var_6959_begin_0, end = var_6959_end_0, end_mask = var_6959_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6959_cast_fp16")]; tensor var_6963_begin_0 = const()[name = tensor("op_6963_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_6963_end_0 = const()[name = tensor("op_6963_end_0"), val = tensor([2, 1280, 1, 144])]; tensor var_6963_end_mask_0 = const()[name = tensor("op_6963_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6963_cast_fp16 = slice_by_index(begin = var_6963_begin_0, end = var_6963_end_0, end_mask = var_6963_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_6963_cast_fp16")]; tensor var_6967_equation_0 = const()[name = tensor("op_6967_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6967_cast_fp16 = einsum(equation = var_6967_equation_0, values = (var_6809_cast_fp16, var_6726_cast_fp16))[name = tensor("op_6967_cast_fp16")]; tensor var_6968_to_fp16 = const()[name = tensor("op_6968_to_fp16"), val = tensor(0x1p-3)]; tensor aw_441_cast_fp16 = mul(x = var_6967_cast_fp16, y = var_6968_to_fp16)[name = tensor("aw_441_cast_fp16")]; tensor var_6971_equation_0 = const()[name = tensor("op_6971_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6971_cast_fp16 = einsum(equation = var_6971_equation_0, values = (var_6813_cast_fp16, var_6730_cast_fp16))[name = tensor("op_6971_cast_fp16")]; tensor var_6972_to_fp16 = const()[name = tensor("op_6972_to_fp16"), val = tensor(0x1p-3)]; tensor aw_443_cast_fp16 = mul(x = var_6971_cast_fp16, y = var_6972_to_fp16)[name = tensor("aw_443_cast_fp16")]; tensor var_6975_equation_0 = const()[name = tensor("op_6975_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6975_cast_fp16 = einsum(equation = var_6975_equation_0, values = (var_6817_cast_fp16, var_6734_cast_fp16))[name = tensor("op_6975_cast_fp16")]; tensor var_6976_to_fp16 = const()[name = tensor("op_6976_to_fp16"), val = tensor(0x1p-3)]; tensor aw_445_cast_fp16 = mul(x = var_6975_cast_fp16, y = var_6976_to_fp16)[name = tensor("aw_445_cast_fp16")]; tensor var_6979_equation_0 = const()[name = tensor("op_6979_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6979_cast_fp16 = einsum(equation = var_6979_equation_0, values = (var_6821_cast_fp16, var_6738_cast_fp16))[name = tensor("op_6979_cast_fp16")]; tensor var_6980_to_fp16 = const()[name = tensor("op_6980_to_fp16"), val = tensor(0x1p-3)]; tensor aw_447_cast_fp16 = mul(x = var_6979_cast_fp16, y = var_6980_to_fp16)[name = tensor("aw_447_cast_fp16")]; tensor var_6983_equation_0 = const()[name = tensor("op_6983_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6983_cast_fp16 = einsum(equation = var_6983_equation_0, values = (var_6825_cast_fp16, var_6742_cast_fp16))[name = tensor("op_6983_cast_fp16")]; tensor var_6984_to_fp16 = const()[name = tensor("op_6984_to_fp16"), val = tensor(0x1p-3)]; tensor aw_449_cast_fp16 = mul(x = var_6983_cast_fp16, y = var_6984_to_fp16)[name = tensor("aw_449_cast_fp16")]; tensor var_6987_equation_0 = const()[name = tensor("op_6987_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6987_cast_fp16 = einsum(equation = var_6987_equation_0, values = (var_6829_cast_fp16, var_6746_cast_fp16))[name = tensor("op_6987_cast_fp16")]; tensor var_6988_to_fp16 = const()[name = tensor("op_6988_to_fp16"), val = tensor(0x1p-3)]; tensor aw_451_cast_fp16 = mul(x = var_6987_cast_fp16, y = var_6988_to_fp16)[name = tensor("aw_451_cast_fp16")]; tensor var_6991_equation_0 = const()[name = tensor("op_6991_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6991_cast_fp16 = einsum(equation = var_6991_equation_0, values = (var_6833_cast_fp16, var_6750_cast_fp16))[name = tensor("op_6991_cast_fp16")]; tensor var_6992_to_fp16 = const()[name = tensor("op_6992_to_fp16"), val = tensor(0x1p-3)]; tensor aw_453_cast_fp16 = mul(x = var_6991_cast_fp16, y = var_6992_to_fp16)[name = tensor("aw_453_cast_fp16")]; tensor var_6995_equation_0 = const()[name = tensor("op_6995_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6995_cast_fp16 = einsum(equation = var_6995_equation_0, values = (var_6837_cast_fp16, var_6754_cast_fp16))[name = tensor("op_6995_cast_fp16")]; tensor var_6996_to_fp16 = const()[name = tensor("op_6996_to_fp16"), val = tensor(0x1p-3)]; tensor aw_455_cast_fp16 = mul(x = var_6995_cast_fp16, y = var_6996_to_fp16)[name = tensor("aw_455_cast_fp16")]; tensor var_6999_equation_0 = const()[name = tensor("op_6999_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6999_cast_fp16 = einsum(equation = var_6999_equation_0, values = (var_6841_cast_fp16, var_6758_cast_fp16))[name = tensor("op_6999_cast_fp16")]; tensor var_7000_to_fp16 = const()[name = tensor("op_7000_to_fp16"), val = tensor(0x1p-3)]; tensor aw_457_cast_fp16 = mul(x = var_6999_cast_fp16, y = var_7000_to_fp16)[name = tensor("aw_457_cast_fp16")]; tensor var_7003_equation_0 = const()[name = tensor("op_7003_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7003_cast_fp16 = einsum(equation = var_7003_equation_0, values = (var_6845_cast_fp16, var_6762_cast_fp16))[name = tensor("op_7003_cast_fp16")]; tensor var_7004_to_fp16 = const()[name = tensor("op_7004_to_fp16"), val = tensor(0x1p-3)]; tensor aw_459_cast_fp16 = mul(x = var_7003_cast_fp16, y = var_7004_to_fp16)[name = tensor("aw_459_cast_fp16")]; tensor var_7007_equation_0 = const()[name = tensor("op_7007_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7007_cast_fp16 = einsum(equation = var_7007_equation_0, values = (var_6849_cast_fp16, var_6766_cast_fp16))[name = tensor("op_7007_cast_fp16")]; tensor var_7008_to_fp16 = const()[name = tensor("op_7008_to_fp16"), val = tensor(0x1p-3)]; tensor aw_461_cast_fp16 = mul(x = var_7007_cast_fp16, y = var_7008_to_fp16)[name = tensor("aw_461_cast_fp16")]; tensor var_7011_equation_0 = const()[name = tensor("op_7011_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7011_cast_fp16 = einsum(equation = var_7011_equation_0, values = (var_6853_cast_fp16, var_6770_cast_fp16))[name = tensor("op_7011_cast_fp16")]; tensor var_7012_to_fp16 = const()[name = tensor("op_7012_to_fp16"), val = tensor(0x1p-3)]; tensor aw_463_cast_fp16 = mul(x = var_7011_cast_fp16, y = var_7012_to_fp16)[name = tensor("aw_463_cast_fp16")]; tensor var_7015_equation_0 = const()[name = tensor("op_7015_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7015_cast_fp16 = einsum(equation = var_7015_equation_0, values = (var_6857_cast_fp16, var_6774_cast_fp16))[name = tensor("op_7015_cast_fp16")]; tensor var_7016_to_fp16 = const()[name = tensor("op_7016_to_fp16"), val = tensor(0x1p-3)]; tensor aw_465_cast_fp16 = mul(x = var_7015_cast_fp16, y = var_7016_to_fp16)[name = tensor("aw_465_cast_fp16")]; tensor var_7019_equation_0 = const()[name = tensor("op_7019_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7019_cast_fp16 = einsum(equation = var_7019_equation_0, values = (var_6861_cast_fp16, var_6778_cast_fp16))[name = tensor("op_7019_cast_fp16")]; tensor var_7020_to_fp16 = const()[name = tensor("op_7020_to_fp16"), val = tensor(0x1p-3)]; tensor aw_467_cast_fp16 = mul(x = var_7019_cast_fp16, y = var_7020_to_fp16)[name = tensor("aw_467_cast_fp16")]; tensor var_7023_equation_0 = const()[name = tensor("op_7023_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7023_cast_fp16 = einsum(equation = var_7023_equation_0, values = (var_6865_cast_fp16, var_6782_cast_fp16))[name = tensor("op_7023_cast_fp16")]; tensor var_7024_to_fp16 = const()[name = tensor("op_7024_to_fp16"), val = tensor(0x1p-3)]; tensor aw_469_cast_fp16 = mul(x = var_7023_cast_fp16, y = var_7024_to_fp16)[name = tensor("aw_469_cast_fp16")]; tensor var_7027_equation_0 = const()[name = tensor("op_7027_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7027_cast_fp16 = einsum(equation = var_7027_equation_0, values = (var_6869_cast_fp16, var_6786_cast_fp16))[name = tensor("op_7027_cast_fp16")]; tensor var_7028_to_fp16 = const()[name = tensor("op_7028_to_fp16"), val = tensor(0x1p-3)]; tensor aw_471_cast_fp16 = mul(x = var_7027_cast_fp16, y = var_7028_to_fp16)[name = tensor("aw_471_cast_fp16")]; tensor var_7031_equation_0 = const()[name = tensor("op_7031_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7031_cast_fp16 = einsum(equation = var_7031_equation_0, values = (var_6873_cast_fp16, var_6790_cast_fp16))[name = tensor("op_7031_cast_fp16")]; tensor var_7032_to_fp16 = const()[name = tensor("op_7032_to_fp16"), val = tensor(0x1p-3)]; tensor aw_473_cast_fp16 = mul(x = var_7031_cast_fp16, y = var_7032_to_fp16)[name = tensor("aw_473_cast_fp16")]; tensor var_7035_equation_0 = const()[name = tensor("op_7035_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7035_cast_fp16 = einsum(equation = var_7035_equation_0, values = (var_6877_cast_fp16, var_6794_cast_fp16))[name = tensor("op_7035_cast_fp16")]; tensor var_7036_to_fp16 = const()[name = tensor("op_7036_to_fp16"), val = tensor(0x1p-3)]; tensor aw_475_cast_fp16 = mul(x = var_7035_cast_fp16, y = var_7036_to_fp16)[name = tensor("aw_475_cast_fp16")]; tensor var_7039_equation_0 = const()[name = tensor("op_7039_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7039_cast_fp16 = einsum(equation = var_7039_equation_0, values = (var_6881_cast_fp16, var_6798_cast_fp16))[name = tensor("op_7039_cast_fp16")]; tensor var_7040_to_fp16 = const()[name = tensor("op_7040_to_fp16"), val = tensor(0x1p-3)]; tensor aw_477_cast_fp16 = mul(x = var_7039_cast_fp16, y = var_7040_to_fp16)[name = tensor("aw_477_cast_fp16")]; tensor var_7043_equation_0 = const()[name = tensor("op_7043_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7043_cast_fp16 = einsum(equation = var_7043_equation_0, values = (var_6885_cast_fp16, var_6802_cast_fp16))[name = tensor("op_7043_cast_fp16")]; tensor var_7044_to_fp16 = const()[name = tensor("op_7044_to_fp16"), val = tensor(0x1p-3)]; tensor aw_479_cast_fp16 = mul(x = var_7043_cast_fp16, y = var_7044_to_fp16)[name = tensor("aw_479_cast_fp16")]; tensor var_7046_cast_fp16 = softmax(axis = var_5599, x = aw_441_cast_fp16)[name = tensor("op_7046_cast_fp16")]; tensor var_7047_cast_fp16 = softmax(axis = var_5599, x = aw_443_cast_fp16)[name = tensor("op_7047_cast_fp16")]; tensor var_7048_cast_fp16 = softmax(axis = var_5599, x = aw_445_cast_fp16)[name = tensor("op_7048_cast_fp16")]; tensor var_7049_cast_fp16 = softmax(axis = var_5599, x = aw_447_cast_fp16)[name = tensor("op_7049_cast_fp16")]; tensor var_7050_cast_fp16 = softmax(axis = var_5599, x = aw_449_cast_fp16)[name = tensor("op_7050_cast_fp16")]; tensor var_7051_cast_fp16 = softmax(axis = var_5599, x = aw_451_cast_fp16)[name = tensor("op_7051_cast_fp16")]; tensor var_7052_cast_fp16 = softmax(axis = var_5599, x = aw_453_cast_fp16)[name = tensor("op_7052_cast_fp16")]; tensor var_7053_cast_fp16 = softmax(axis = var_5599, x = aw_455_cast_fp16)[name = tensor("op_7053_cast_fp16")]; tensor var_7054_cast_fp16 = softmax(axis = var_5599, x = aw_457_cast_fp16)[name = tensor("op_7054_cast_fp16")]; tensor var_7055_cast_fp16 = softmax(axis = var_5599, x = aw_459_cast_fp16)[name = tensor("op_7055_cast_fp16")]; tensor var_7056_cast_fp16 = softmax(axis = var_5599, x = aw_461_cast_fp16)[name = tensor("op_7056_cast_fp16")]; tensor var_7057_cast_fp16 = softmax(axis = var_5599, x = aw_463_cast_fp16)[name = tensor("op_7057_cast_fp16")]; tensor var_7058_cast_fp16 = softmax(axis = var_5599, x = aw_465_cast_fp16)[name = tensor("op_7058_cast_fp16")]; tensor var_7059_cast_fp16 = softmax(axis = var_5599, x = aw_467_cast_fp16)[name = tensor("op_7059_cast_fp16")]; tensor var_7060_cast_fp16 = softmax(axis = var_5599, x = aw_469_cast_fp16)[name = tensor("op_7060_cast_fp16")]; tensor var_7061_cast_fp16 = softmax(axis = var_5599, x = aw_471_cast_fp16)[name = tensor("op_7061_cast_fp16")]; tensor var_7062_cast_fp16 = softmax(axis = var_5599, x = aw_473_cast_fp16)[name = tensor("op_7062_cast_fp16")]; tensor var_7063_cast_fp16 = softmax(axis = var_5599, x = aw_475_cast_fp16)[name = tensor("op_7063_cast_fp16")]; tensor var_7064_cast_fp16 = softmax(axis = var_5599, x = aw_477_cast_fp16)[name = tensor("op_7064_cast_fp16")]; tensor var_7065_cast_fp16 = softmax(axis = var_5599, x = aw_479_cast_fp16)[name = tensor("op_7065_cast_fp16")]; tensor var_7067_equation_0 = const()[name = tensor("op_7067_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7067_cast_fp16 = einsum(equation = var_7067_equation_0, values = (var_6887_cast_fp16, var_7046_cast_fp16))[name = tensor("op_7067_cast_fp16")]; tensor var_7069_equation_0 = const()[name = tensor("op_7069_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7069_cast_fp16 = einsum(equation = var_7069_equation_0, values = (var_6891_cast_fp16, var_7047_cast_fp16))[name = tensor("op_7069_cast_fp16")]; tensor var_7071_equation_0 = const()[name = tensor("op_7071_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7071_cast_fp16 = einsum(equation = var_7071_equation_0, values = (var_6895_cast_fp16, var_7048_cast_fp16))[name = tensor("op_7071_cast_fp16")]; tensor var_7073_equation_0 = const()[name = tensor("op_7073_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7073_cast_fp16 = einsum(equation = var_7073_equation_0, values = (var_6899_cast_fp16, var_7049_cast_fp16))[name = tensor("op_7073_cast_fp16")]; tensor var_7075_equation_0 = const()[name = tensor("op_7075_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7075_cast_fp16 = einsum(equation = var_7075_equation_0, values = (var_6903_cast_fp16, var_7050_cast_fp16))[name = tensor("op_7075_cast_fp16")]; tensor var_7077_equation_0 = const()[name = tensor("op_7077_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7077_cast_fp16 = einsum(equation = var_7077_equation_0, values = (var_6907_cast_fp16, var_7051_cast_fp16))[name = tensor("op_7077_cast_fp16")]; tensor var_7079_equation_0 = const()[name = tensor("op_7079_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7079_cast_fp16 = einsum(equation = var_7079_equation_0, values = (var_6911_cast_fp16, var_7052_cast_fp16))[name = tensor("op_7079_cast_fp16")]; tensor var_7081_equation_0 = const()[name = tensor("op_7081_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7081_cast_fp16 = einsum(equation = var_7081_equation_0, values = (var_6915_cast_fp16, var_7053_cast_fp16))[name = tensor("op_7081_cast_fp16")]; tensor var_7083_equation_0 = const()[name = tensor("op_7083_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7083_cast_fp16 = einsum(equation = var_7083_equation_0, values = (var_6919_cast_fp16, var_7054_cast_fp16))[name = tensor("op_7083_cast_fp16")]; tensor var_7085_equation_0 = const()[name = tensor("op_7085_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7085_cast_fp16 = einsum(equation = var_7085_equation_0, values = (var_6923_cast_fp16, var_7055_cast_fp16))[name = tensor("op_7085_cast_fp16")]; tensor var_7087_equation_0 = const()[name = tensor("op_7087_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7087_cast_fp16 = einsum(equation = var_7087_equation_0, values = (var_6927_cast_fp16, var_7056_cast_fp16))[name = tensor("op_7087_cast_fp16")]; tensor var_7089_equation_0 = const()[name = tensor("op_7089_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7089_cast_fp16 = einsum(equation = var_7089_equation_0, values = (var_6931_cast_fp16, var_7057_cast_fp16))[name = tensor("op_7089_cast_fp16")]; tensor var_7091_equation_0 = const()[name = tensor("op_7091_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7091_cast_fp16 = einsum(equation = var_7091_equation_0, values = (var_6935_cast_fp16, var_7058_cast_fp16))[name = tensor("op_7091_cast_fp16")]; tensor var_7093_equation_0 = const()[name = tensor("op_7093_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7093_cast_fp16 = einsum(equation = var_7093_equation_0, values = (var_6939_cast_fp16, var_7059_cast_fp16))[name = tensor("op_7093_cast_fp16")]; tensor var_7095_equation_0 = const()[name = tensor("op_7095_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7095_cast_fp16 = einsum(equation = var_7095_equation_0, values = (var_6943_cast_fp16, var_7060_cast_fp16))[name = tensor("op_7095_cast_fp16")]; tensor var_7097_equation_0 = const()[name = tensor("op_7097_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7097_cast_fp16 = einsum(equation = var_7097_equation_0, values = (var_6947_cast_fp16, var_7061_cast_fp16))[name = tensor("op_7097_cast_fp16")]; tensor var_7099_equation_0 = const()[name = tensor("op_7099_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7099_cast_fp16 = einsum(equation = var_7099_equation_0, values = (var_6951_cast_fp16, var_7062_cast_fp16))[name = tensor("op_7099_cast_fp16")]; tensor var_7101_equation_0 = const()[name = tensor("op_7101_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7101_cast_fp16 = einsum(equation = var_7101_equation_0, values = (var_6955_cast_fp16, var_7063_cast_fp16))[name = tensor("op_7101_cast_fp16")]; tensor var_7103_equation_0 = const()[name = tensor("op_7103_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7103_cast_fp16 = einsum(equation = var_7103_equation_0, values = (var_6959_cast_fp16, var_7064_cast_fp16))[name = tensor("op_7103_cast_fp16")]; tensor var_7105_equation_0 = const()[name = tensor("op_7105_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7105_cast_fp16 = einsum(equation = var_7105_equation_0, values = (var_6963_cast_fp16, var_7065_cast_fp16))[name = tensor("op_7105_cast_fp16")]; tensor input_327_interleave_0 = const()[name = tensor("input_327_interleave_0"), val = tensor(false)]; tensor input_327_cast_fp16 = concat(axis = var_5599, interleave = input_327_interleave_0, values = (var_7067_cast_fp16, var_7069_cast_fp16, var_7071_cast_fp16, var_7073_cast_fp16, var_7075_cast_fp16, var_7077_cast_fp16, var_7079_cast_fp16, var_7081_cast_fp16, var_7083_cast_fp16, var_7085_cast_fp16, var_7087_cast_fp16, var_7089_cast_fp16, var_7091_cast_fp16, var_7093_cast_fp16, var_7095_cast_fp16, var_7097_cast_fp16, var_7099_cast_fp16, var_7101_cast_fp16, var_7103_cast_fp16, var_7105_cast_fp16))[name = tensor("input_327_cast_fp16")]; tensor var_7111 = const()[name = tensor("op_7111"), val = tensor([1, 1])]; tensor var_7113 = const()[name = tensor("op_7113"), val = tensor([1, 1])]; tensor var_7115_pad_type_0 = const()[name = tensor("op_7115_pad_type_0"), val = tensor("custom")]; tensor var_7115_pad_0 = const()[name = tensor("op_7115_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(415015936)))]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418292800)))]; tensor var_7115_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_7113, groups = var_5599, pad = var_7115_pad_0, pad_type = var_7115_pad_type_0, strides = var_7111, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_327_cast_fp16)[name = tensor("op_7115_cast_fp16")]; tensor inputs_51_cast_fp16 = add(x = var_7115_cast_fp16, y = inputs_49_cast_fp16)[name = tensor("inputs_51_cast_fp16")]; tensor var_7119 = const()[name = tensor("op_7119"), val = tensor([1])]; tensor channels_mean_51_cast_fp16 = reduce_mean(axes = var_7119, keep_dims = var_5594, x = inputs_51_cast_fp16)[name = tensor("channels_mean_51_cast_fp16")]; tensor zero_mean_51_cast_fp16 = sub(x = inputs_51_cast_fp16, y = channels_mean_51_cast_fp16)[name = tensor("zero_mean_51_cast_fp16")]; tensor zero_mean_sq_51_cast_fp16 = mul(x = zero_mean_51_cast_fp16, y = zero_mean_51_cast_fp16)[name = tensor("zero_mean_sq_51_cast_fp16")]; tensor var_7123 = const()[name = tensor("op_7123"), val = tensor([1])]; tensor var_7124_cast_fp16 = reduce_mean(axes = var_7123, keep_dims = var_5594, x = zero_mean_sq_51_cast_fp16)[name = tensor("op_7124_cast_fp16")]; tensor var_7125_to_fp16 = const()[name = tensor("op_7125_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_7126_cast_fp16 = add(x = var_7124_cast_fp16, y = var_7125_to_fp16)[name = tensor("op_7126_cast_fp16")]; tensor denom_51_epsilon_0_to_fp16 = const()[name = tensor("denom_51_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_51_cast_fp16 = rsqrt(epsilon = denom_51_epsilon_0_to_fp16, x = var_7126_cast_fp16)[name = tensor("denom_51_cast_fp16")]; tensor out_51_cast_fp16 = mul(x = zero_mean_51_cast_fp16, y = denom_51_cast_fp16)[name = tensor("out_51_cast_fp16")]; tensor var_7130_to_fp16 = const()[name = tensor("op_7130_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418295424)))]; tensor var_7131_cast_fp16 = add(x = out_51_cast_fp16, y = var_7130_to_fp16)[name = tensor("op_7131_cast_fp16")]; tensor var_7133_to_fp16 = const()[name = tensor("op_7133_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418298048)))]; tensor hidden_states_189_cast_fp16 = mul(x = var_7131_cast_fp16, y = var_7133_to_fp16)[name = tensor("hidden_states_189_cast_fp16")]; tensor var_7140 = const()[name = tensor("op_7140"), val = tensor([1, 1])]; tensor var_7142 = const()[name = tensor("op_7142"), val = tensor([1, 1])]; tensor q_35_pad_type_0 = const()[name = tensor("q_35_pad_type_0"), val = tensor("custom")]; tensor q_35_pad_0 = const()[name = tensor("q_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418300672)))]; tensor q_35_cast_fp16 = conv(dilations = var_7142, groups = var_5599, pad = q_35_pad_0, pad_type = q_35_pad_type_0, strides = var_7140, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_189_cast_fp16)[name = tensor("q_35_cast_fp16")]; tensor var_7146 = const()[name = tensor("op_7146"), val = tensor([1, 1])]; tensor var_7148 = const()[name = tensor("op_7148"), val = tensor([1, 1])]; tensor k_69_pad_type_0 = const()[name = tensor("k_69_pad_type_0"), val = tensor("custom")]; tensor k_69_pad_0 = const()[name = tensor("k_69_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421577536)))]; tensor k_69_cast_fp16 = conv(dilations = var_7148, groups = var_5599, pad = k_69_pad_0, pad_type = k_69_pad_type_0, strides = var_7146, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_69_cast_fp16")]; tensor var_7152 = const()[name = tensor("op_7152"), val = tensor([1, 1])]; tensor var_7154 = const()[name = tensor("op_7154"), val = tensor([1, 1])]; tensor v_35_pad_type_0 = const()[name = tensor("v_35_pad_type_0"), val = tensor("custom")]; tensor v_35_pad_0 = const()[name = tensor("v_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(424199040)))]; tensor v_35_cast_fp16 = conv(dilations = var_7154, groups = var_5599, pad = v_35_pad_0, pad_type = v_35_pad_type_0, strides = var_7152, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_35_cast_fp16")]; tensor var_7158_begin_0 = const()[name = tensor("op_7158_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7158_end_0 = const()[name = tensor("op_7158_end_0"), val = tensor([2, 64, 1, 144])]; tensor var_7158_end_mask_0 = const()[name = tensor("op_7158_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7158_cast_fp16 = slice_by_index(begin = var_7158_begin_0, end = var_7158_end_0, end_mask = var_7158_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7158_cast_fp16")]; tensor var_7162_begin_0 = const()[name = tensor("op_7162_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_7162_end_0 = const()[name = tensor("op_7162_end_0"), val = tensor([2, 128, 1, 144])]; tensor var_7162_end_mask_0 = const()[name = tensor("op_7162_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7162_cast_fp16 = slice_by_index(begin = var_7162_begin_0, end = var_7162_end_0, end_mask = var_7162_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7162_cast_fp16")]; tensor var_7166_begin_0 = const()[name = tensor("op_7166_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_7166_end_0 = const()[name = tensor("op_7166_end_0"), val = tensor([2, 192, 1, 144])]; tensor var_7166_end_mask_0 = const()[name = tensor("op_7166_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7166_cast_fp16 = slice_by_index(begin = var_7166_begin_0, end = var_7166_end_0, end_mask = var_7166_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7166_cast_fp16")]; tensor var_7170_begin_0 = const()[name = tensor("op_7170_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_7170_end_0 = const()[name = tensor("op_7170_end_0"), val = tensor([2, 256, 1, 144])]; tensor var_7170_end_mask_0 = const()[name = tensor("op_7170_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7170_cast_fp16 = slice_by_index(begin = var_7170_begin_0, end = var_7170_end_0, end_mask = var_7170_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7170_cast_fp16")]; tensor var_7174_begin_0 = const()[name = tensor("op_7174_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_7174_end_0 = const()[name = tensor("op_7174_end_0"), val = tensor([2, 320, 1, 144])]; tensor var_7174_end_mask_0 = const()[name = tensor("op_7174_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7174_cast_fp16 = slice_by_index(begin = var_7174_begin_0, end = var_7174_end_0, end_mask = var_7174_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7174_cast_fp16")]; tensor var_7178_begin_0 = const()[name = tensor("op_7178_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7178_end_0 = const()[name = tensor("op_7178_end_0"), val = tensor([2, 384, 1, 144])]; tensor var_7178_end_mask_0 = const()[name = tensor("op_7178_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7178_cast_fp16 = slice_by_index(begin = var_7178_begin_0, end = var_7178_end_0, end_mask = var_7178_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7178_cast_fp16")]; tensor var_7182_begin_0 = const()[name = tensor("op_7182_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_7182_end_0 = const()[name = tensor("op_7182_end_0"), val = tensor([2, 448, 1, 144])]; tensor var_7182_end_mask_0 = const()[name = tensor("op_7182_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7182_cast_fp16 = slice_by_index(begin = var_7182_begin_0, end = var_7182_end_0, end_mask = var_7182_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7182_cast_fp16")]; tensor var_7186_begin_0 = const()[name = tensor("op_7186_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_7186_end_0 = const()[name = tensor("op_7186_end_0"), val = tensor([2, 512, 1, 144])]; tensor var_7186_end_mask_0 = const()[name = tensor("op_7186_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7186_cast_fp16 = slice_by_index(begin = var_7186_begin_0, end = var_7186_end_0, end_mask = var_7186_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7186_cast_fp16")]; tensor var_7190_begin_0 = const()[name = tensor("op_7190_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_7190_end_0 = const()[name = tensor("op_7190_end_0"), val = tensor([2, 576, 1, 144])]; tensor var_7190_end_mask_0 = const()[name = tensor("op_7190_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7190_cast_fp16 = slice_by_index(begin = var_7190_begin_0, end = var_7190_end_0, end_mask = var_7190_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7190_cast_fp16")]; tensor var_7194_begin_0 = const()[name = tensor("op_7194_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_7194_end_0 = const()[name = tensor("op_7194_end_0"), val = tensor([2, 640, 1, 144])]; tensor var_7194_end_mask_0 = const()[name = tensor("op_7194_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7194_cast_fp16 = slice_by_index(begin = var_7194_begin_0, end = var_7194_end_0, end_mask = var_7194_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7194_cast_fp16")]; tensor var_7198_begin_0 = const()[name = tensor("op_7198_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7198_end_0 = const()[name = tensor("op_7198_end_0"), val = tensor([2, 704, 1, 144])]; tensor var_7198_end_mask_0 = const()[name = tensor("op_7198_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7198_cast_fp16 = slice_by_index(begin = var_7198_begin_0, end = var_7198_end_0, end_mask = var_7198_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7198_cast_fp16")]; tensor var_7202_begin_0 = const()[name = tensor("op_7202_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_7202_end_0 = const()[name = tensor("op_7202_end_0"), val = tensor([2, 768, 1, 144])]; tensor var_7202_end_mask_0 = const()[name = tensor("op_7202_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7202_cast_fp16 = slice_by_index(begin = var_7202_begin_0, end = var_7202_end_0, end_mask = var_7202_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7202_cast_fp16")]; tensor var_7206_begin_0 = const()[name = tensor("op_7206_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_7206_end_0 = const()[name = tensor("op_7206_end_0"), val = tensor([2, 832, 1, 144])]; tensor var_7206_end_mask_0 = const()[name = tensor("op_7206_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7206_cast_fp16 = slice_by_index(begin = var_7206_begin_0, end = var_7206_end_0, end_mask = var_7206_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7206_cast_fp16")]; tensor var_7210_begin_0 = const()[name = tensor("op_7210_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_7210_end_0 = const()[name = tensor("op_7210_end_0"), val = tensor([2, 896, 1, 144])]; tensor var_7210_end_mask_0 = const()[name = tensor("op_7210_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7210_cast_fp16 = slice_by_index(begin = var_7210_begin_0, end = var_7210_end_0, end_mask = var_7210_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7210_cast_fp16")]; tensor var_7214_begin_0 = const()[name = tensor("op_7214_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_7214_end_0 = const()[name = tensor("op_7214_end_0"), val = tensor([2, 960, 1, 144])]; tensor var_7214_end_mask_0 = const()[name = tensor("op_7214_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7214_cast_fp16 = slice_by_index(begin = var_7214_begin_0, end = var_7214_end_0, end_mask = var_7214_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7214_cast_fp16")]; tensor var_7218_begin_0 = const()[name = tensor("op_7218_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_7218_end_0 = const()[name = tensor("op_7218_end_0"), val = tensor([2, 1024, 1, 144])]; tensor var_7218_end_mask_0 = const()[name = tensor("op_7218_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7218_cast_fp16 = slice_by_index(begin = var_7218_begin_0, end = var_7218_end_0, end_mask = var_7218_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7218_cast_fp16")]; tensor var_7222_begin_0 = const()[name = tensor("op_7222_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_7222_end_0 = const()[name = tensor("op_7222_end_0"), val = tensor([2, 1088, 1, 144])]; tensor var_7222_end_mask_0 = const()[name = tensor("op_7222_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7222_cast_fp16 = slice_by_index(begin = var_7222_begin_0, end = var_7222_end_0, end_mask = var_7222_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7222_cast_fp16")]; tensor var_7226_begin_0 = const()[name = tensor("op_7226_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_7226_end_0 = const()[name = tensor("op_7226_end_0"), val = tensor([2, 1152, 1, 144])]; tensor var_7226_end_mask_0 = const()[name = tensor("op_7226_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7226_cast_fp16 = slice_by_index(begin = var_7226_begin_0, end = var_7226_end_0, end_mask = var_7226_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7226_cast_fp16")]; tensor var_7230_begin_0 = const()[name = tensor("op_7230_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_7230_end_0 = const()[name = tensor("op_7230_end_0"), val = tensor([2, 1216, 1, 144])]; tensor var_7230_end_mask_0 = const()[name = tensor("op_7230_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7230_cast_fp16 = slice_by_index(begin = var_7230_begin_0, end = var_7230_end_0, end_mask = var_7230_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7230_cast_fp16")]; tensor var_7234_begin_0 = const()[name = tensor("op_7234_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_7234_end_0 = const()[name = tensor("op_7234_end_0"), val = tensor([2, 1280, 1, 144])]; tensor var_7234_end_mask_0 = const()[name = tensor("op_7234_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7234_cast_fp16 = slice_by_index(begin = var_7234_begin_0, end = var_7234_end_0, end_mask = var_7234_end_mask_0, x = q_35_cast_fp16)[name = tensor("op_7234_cast_fp16")]; tensor k_71_perm_0 = const()[name = tensor("k_71_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_7241_begin_0 = const()[name = tensor("op_7241_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7241_end_0 = const()[name = tensor("op_7241_end_0"), val = tensor([2, 77, 1, 64])]; tensor var_7241_end_mask_0 = const()[name = tensor("op_7241_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_14 = transpose(perm = k_71_perm_0, x = k_69_cast_fp16)[name = tensor("transpose_14")]; tensor var_7241_cast_fp16 = slice_by_index(begin = var_7241_begin_0, end = var_7241_end_0, end_mask = var_7241_end_mask_0, x = transpose_14)[name = tensor("op_7241_cast_fp16")]; tensor var_7245_begin_0 = const()[name = tensor("op_7245_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_7245_end_0 = const()[name = tensor("op_7245_end_0"), val = tensor([2, 77, 1, 128])]; tensor var_7245_end_mask_0 = const()[name = tensor("op_7245_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7245_cast_fp16 = slice_by_index(begin = var_7245_begin_0, end = var_7245_end_0, end_mask = var_7245_end_mask_0, x = transpose_14)[name = tensor("op_7245_cast_fp16")]; tensor var_7249_begin_0 = const()[name = tensor("op_7249_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_7249_end_0 = const()[name = tensor("op_7249_end_0"), val = tensor([2, 77, 1, 192])]; tensor var_7249_end_mask_0 = const()[name = tensor("op_7249_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7249_cast_fp16 = slice_by_index(begin = var_7249_begin_0, end = var_7249_end_0, end_mask = var_7249_end_mask_0, x = transpose_14)[name = tensor("op_7249_cast_fp16")]; tensor var_7253_begin_0 = const()[name = tensor("op_7253_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_7253_end_0 = const()[name = tensor("op_7253_end_0"), val = tensor([2, 77, 1, 256])]; tensor var_7253_end_mask_0 = const()[name = tensor("op_7253_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7253_cast_fp16 = slice_by_index(begin = var_7253_begin_0, end = var_7253_end_0, end_mask = var_7253_end_mask_0, x = transpose_14)[name = tensor("op_7253_cast_fp16")]; tensor var_7257_begin_0 = const()[name = tensor("op_7257_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_7257_end_0 = const()[name = tensor("op_7257_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_7257_end_mask_0 = const()[name = tensor("op_7257_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7257_cast_fp16 = slice_by_index(begin = var_7257_begin_0, end = var_7257_end_0, end_mask = var_7257_end_mask_0, x = transpose_14)[name = tensor("op_7257_cast_fp16")]; tensor var_7261_begin_0 = const()[name = tensor("op_7261_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_7261_end_0 = const()[name = tensor("op_7261_end_0"), val = tensor([2, 77, 1, 384])]; tensor var_7261_end_mask_0 = const()[name = tensor("op_7261_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7261_cast_fp16 = slice_by_index(begin = var_7261_begin_0, end = var_7261_end_0, end_mask = var_7261_end_mask_0, x = transpose_14)[name = tensor("op_7261_cast_fp16")]; tensor var_7265_begin_0 = const()[name = tensor("op_7265_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_7265_end_0 = const()[name = tensor("op_7265_end_0"), val = tensor([2, 77, 1, 448])]; tensor var_7265_end_mask_0 = const()[name = tensor("op_7265_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7265_cast_fp16 = slice_by_index(begin = var_7265_begin_0, end = var_7265_end_0, end_mask = var_7265_end_mask_0, x = transpose_14)[name = tensor("op_7265_cast_fp16")]; tensor var_7269_begin_0 = const()[name = tensor("op_7269_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_7269_end_0 = const()[name = tensor("op_7269_end_0"), val = tensor([2, 77, 1, 512])]; tensor var_7269_end_mask_0 = const()[name = tensor("op_7269_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7269_cast_fp16 = slice_by_index(begin = var_7269_begin_0, end = var_7269_end_0, end_mask = var_7269_end_mask_0, x = transpose_14)[name = tensor("op_7269_cast_fp16")]; tensor var_7273_begin_0 = const()[name = tensor("op_7273_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7273_end_0 = const()[name = tensor("op_7273_end_0"), val = tensor([2, 77, 1, 576])]; tensor var_7273_end_mask_0 = const()[name = tensor("op_7273_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7273_cast_fp16 = slice_by_index(begin = var_7273_begin_0, end = var_7273_end_0, end_mask = var_7273_end_mask_0, x = transpose_14)[name = tensor("op_7273_cast_fp16")]; tensor var_7277_begin_0 = const()[name = tensor("op_7277_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_7277_end_0 = const()[name = tensor("op_7277_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_7277_end_mask_0 = const()[name = tensor("op_7277_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7277_cast_fp16 = slice_by_index(begin = var_7277_begin_0, end = var_7277_end_0, end_mask = var_7277_end_mask_0, x = transpose_14)[name = tensor("op_7277_cast_fp16")]; tensor var_7281_begin_0 = const()[name = tensor("op_7281_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_7281_end_0 = const()[name = tensor("op_7281_end_0"), val = tensor([2, 77, 1, 704])]; tensor var_7281_end_mask_0 = const()[name = tensor("op_7281_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7281_cast_fp16 = slice_by_index(begin = var_7281_begin_0, end = var_7281_end_0, end_mask = var_7281_end_mask_0, x = transpose_14)[name = tensor("op_7281_cast_fp16")]; tensor var_7285_begin_0 = const()[name = tensor("op_7285_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_7285_end_0 = const()[name = tensor("op_7285_end_0"), val = tensor([2, 77, 1, 768])]; tensor var_7285_end_mask_0 = const()[name = tensor("op_7285_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7285_cast_fp16 = slice_by_index(begin = var_7285_begin_0, end = var_7285_end_0, end_mask = var_7285_end_mask_0, x = transpose_14)[name = tensor("op_7285_cast_fp16")]; tensor var_7289_begin_0 = const()[name = tensor("op_7289_begin_0"), val = tensor([0, 0, 0, 768])]; tensor var_7289_end_0 = const()[name = tensor("op_7289_end_0"), val = tensor([2, 77, 1, 832])]; tensor var_7289_end_mask_0 = const()[name = tensor("op_7289_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7289_cast_fp16 = slice_by_index(begin = var_7289_begin_0, end = var_7289_end_0, end_mask = var_7289_end_mask_0, x = transpose_14)[name = tensor("op_7289_cast_fp16")]; tensor var_7293_begin_0 = const()[name = tensor("op_7293_begin_0"), val = tensor([0, 0, 0, 832])]; tensor var_7293_end_0 = const()[name = tensor("op_7293_end_0"), val = tensor([2, 77, 1, 896])]; tensor var_7293_end_mask_0 = const()[name = tensor("op_7293_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7293_cast_fp16 = slice_by_index(begin = var_7293_begin_0, end = var_7293_end_0, end_mask = var_7293_end_mask_0, x = transpose_14)[name = tensor("op_7293_cast_fp16")]; tensor var_7297_begin_0 = const()[name = tensor("op_7297_begin_0"), val = tensor([0, 0, 0, 896])]; tensor var_7297_end_0 = const()[name = tensor("op_7297_end_0"), val = tensor([2, 77, 1, 960])]; tensor var_7297_end_mask_0 = const()[name = tensor("op_7297_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7297_cast_fp16 = slice_by_index(begin = var_7297_begin_0, end = var_7297_end_0, end_mask = var_7297_end_mask_0, x = transpose_14)[name = tensor("op_7297_cast_fp16")]; tensor var_7301_begin_0 = const()[name = tensor("op_7301_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_7301_end_0 = const()[name = tensor("op_7301_end_0"), val = tensor([2, 77, 1, 1024])]; tensor var_7301_end_mask_0 = const()[name = tensor("op_7301_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7301_cast_fp16 = slice_by_index(begin = var_7301_begin_0, end = var_7301_end_0, end_mask = var_7301_end_mask_0, x = transpose_14)[name = tensor("op_7301_cast_fp16")]; tensor var_7305_begin_0 = const()[name = tensor("op_7305_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7305_end_0 = const()[name = tensor("op_7305_end_0"), val = tensor([2, 77, 1, 1088])]; tensor var_7305_end_mask_0 = const()[name = tensor("op_7305_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7305_cast_fp16 = slice_by_index(begin = var_7305_begin_0, end = var_7305_end_0, end_mask = var_7305_end_mask_0, x = transpose_14)[name = tensor("op_7305_cast_fp16")]; tensor var_7309_begin_0 = const()[name = tensor("op_7309_begin_0"), val = tensor([0, 0, 0, 1088])]; tensor var_7309_end_0 = const()[name = tensor("op_7309_end_0"), val = tensor([2, 77, 1, 1152])]; tensor var_7309_end_mask_0 = const()[name = tensor("op_7309_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7309_cast_fp16 = slice_by_index(begin = var_7309_begin_0, end = var_7309_end_0, end_mask = var_7309_end_mask_0, x = transpose_14)[name = tensor("op_7309_cast_fp16")]; tensor var_7313_begin_0 = const()[name = tensor("op_7313_begin_0"), val = tensor([0, 0, 0, 1152])]; tensor var_7313_end_0 = const()[name = tensor("op_7313_end_0"), val = tensor([2, 77, 1, 1216])]; tensor var_7313_end_mask_0 = const()[name = tensor("op_7313_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7313_cast_fp16 = slice_by_index(begin = var_7313_begin_0, end = var_7313_end_0, end_mask = var_7313_end_mask_0, x = transpose_14)[name = tensor("op_7313_cast_fp16")]; tensor var_7317_begin_0 = const()[name = tensor("op_7317_begin_0"), val = tensor([0, 0, 0, 1216])]; tensor var_7317_end_0 = const()[name = tensor("op_7317_end_0"), val = tensor([2, 77, 1, 1280])]; tensor var_7317_end_mask_0 = const()[name = tensor("op_7317_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7317_cast_fp16 = slice_by_index(begin = var_7317_begin_0, end = var_7317_end_0, end_mask = var_7317_end_mask_0, x = transpose_14)[name = tensor("op_7317_cast_fp16")]; tensor var_7319_begin_0 = const()[name = tensor("op_7319_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7319_end_0 = const()[name = tensor("op_7319_end_0"), val = tensor([2, 64, 1, 77])]; tensor var_7319_end_mask_0 = const()[name = tensor("op_7319_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7319_cast_fp16 = slice_by_index(begin = var_7319_begin_0, end = var_7319_end_0, end_mask = var_7319_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7319_cast_fp16")]; tensor var_7323_begin_0 = const()[name = tensor("op_7323_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_7323_end_0 = const()[name = tensor("op_7323_end_0"), val = tensor([2, 128, 1, 77])]; tensor var_7323_end_mask_0 = const()[name = tensor("op_7323_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7323_cast_fp16 = slice_by_index(begin = var_7323_begin_0, end = var_7323_end_0, end_mask = var_7323_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7323_cast_fp16")]; tensor var_7327_begin_0 = const()[name = tensor("op_7327_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_7327_end_0 = const()[name = tensor("op_7327_end_0"), val = tensor([2, 192, 1, 77])]; tensor var_7327_end_mask_0 = const()[name = tensor("op_7327_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7327_cast_fp16 = slice_by_index(begin = var_7327_begin_0, end = var_7327_end_0, end_mask = var_7327_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7327_cast_fp16")]; tensor var_7331_begin_0 = const()[name = tensor("op_7331_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_7331_end_0 = const()[name = tensor("op_7331_end_0"), val = tensor([2, 256, 1, 77])]; tensor var_7331_end_mask_0 = const()[name = tensor("op_7331_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7331_cast_fp16 = slice_by_index(begin = var_7331_begin_0, end = var_7331_end_0, end_mask = var_7331_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7331_cast_fp16")]; tensor var_7335_begin_0 = const()[name = tensor("op_7335_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_7335_end_0 = const()[name = tensor("op_7335_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_7335_end_mask_0 = const()[name = tensor("op_7335_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7335_cast_fp16 = slice_by_index(begin = var_7335_begin_0, end = var_7335_end_0, end_mask = var_7335_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7335_cast_fp16")]; tensor var_7339_begin_0 = const()[name = tensor("op_7339_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7339_end_0 = const()[name = tensor("op_7339_end_0"), val = tensor([2, 384, 1, 77])]; tensor var_7339_end_mask_0 = const()[name = tensor("op_7339_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7339_cast_fp16 = slice_by_index(begin = var_7339_begin_0, end = var_7339_end_0, end_mask = var_7339_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7339_cast_fp16")]; tensor var_7343_begin_0 = const()[name = tensor("op_7343_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_7343_end_0 = const()[name = tensor("op_7343_end_0"), val = tensor([2, 448, 1, 77])]; tensor var_7343_end_mask_0 = const()[name = tensor("op_7343_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7343_cast_fp16 = slice_by_index(begin = var_7343_begin_0, end = var_7343_end_0, end_mask = var_7343_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7343_cast_fp16")]; tensor var_7347_begin_0 = const()[name = tensor("op_7347_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_7347_end_0 = const()[name = tensor("op_7347_end_0"), val = tensor([2, 512, 1, 77])]; tensor var_7347_end_mask_0 = const()[name = tensor("op_7347_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7347_cast_fp16 = slice_by_index(begin = var_7347_begin_0, end = var_7347_end_0, end_mask = var_7347_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7347_cast_fp16")]; tensor var_7351_begin_0 = const()[name = tensor("op_7351_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_7351_end_0 = const()[name = tensor("op_7351_end_0"), val = tensor([2, 576, 1, 77])]; tensor var_7351_end_mask_0 = const()[name = tensor("op_7351_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7351_cast_fp16 = slice_by_index(begin = var_7351_begin_0, end = var_7351_end_0, end_mask = var_7351_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7351_cast_fp16")]; tensor var_7355_begin_0 = const()[name = tensor("op_7355_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_7355_end_0 = const()[name = tensor("op_7355_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_7355_end_mask_0 = const()[name = tensor("op_7355_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7355_cast_fp16 = slice_by_index(begin = var_7355_begin_0, end = var_7355_end_0, end_mask = var_7355_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7355_cast_fp16")]; tensor var_7359_begin_0 = const()[name = tensor("op_7359_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7359_end_0 = const()[name = tensor("op_7359_end_0"), val = tensor([2, 704, 1, 77])]; tensor var_7359_end_mask_0 = const()[name = tensor("op_7359_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7359_cast_fp16 = slice_by_index(begin = var_7359_begin_0, end = var_7359_end_0, end_mask = var_7359_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7359_cast_fp16")]; tensor var_7363_begin_0 = const()[name = tensor("op_7363_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_7363_end_0 = const()[name = tensor("op_7363_end_0"), val = tensor([2, 768, 1, 77])]; tensor var_7363_end_mask_0 = const()[name = tensor("op_7363_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7363_cast_fp16 = slice_by_index(begin = var_7363_begin_0, end = var_7363_end_0, end_mask = var_7363_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7363_cast_fp16")]; tensor var_7367_begin_0 = const()[name = tensor("op_7367_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_7367_end_0 = const()[name = tensor("op_7367_end_0"), val = tensor([2, 832, 1, 77])]; tensor var_7367_end_mask_0 = const()[name = tensor("op_7367_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7367_cast_fp16 = slice_by_index(begin = var_7367_begin_0, end = var_7367_end_0, end_mask = var_7367_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7367_cast_fp16")]; tensor var_7371_begin_0 = const()[name = tensor("op_7371_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_7371_end_0 = const()[name = tensor("op_7371_end_0"), val = tensor([2, 896, 1, 77])]; tensor var_7371_end_mask_0 = const()[name = tensor("op_7371_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7371_cast_fp16 = slice_by_index(begin = var_7371_begin_0, end = var_7371_end_0, end_mask = var_7371_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7371_cast_fp16")]; tensor var_7375_begin_0 = const()[name = tensor("op_7375_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_7375_end_0 = const()[name = tensor("op_7375_end_0"), val = tensor([2, 960, 1, 77])]; tensor var_7375_end_mask_0 = const()[name = tensor("op_7375_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7375_cast_fp16 = slice_by_index(begin = var_7375_begin_0, end = var_7375_end_0, end_mask = var_7375_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7375_cast_fp16")]; tensor var_7379_begin_0 = const()[name = tensor("op_7379_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_7379_end_0 = const()[name = tensor("op_7379_end_0"), val = tensor([2, 1024, 1, 77])]; tensor var_7379_end_mask_0 = const()[name = tensor("op_7379_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7379_cast_fp16 = slice_by_index(begin = var_7379_begin_0, end = var_7379_end_0, end_mask = var_7379_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7379_cast_fp16")]; tensor var_7383_begin_0 = const()[name = tensor("op_7383_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_7383_end_0 = const()[name = tensor("op_7383_end_0"), val = tensor([2, 1088, 1, 77])]; tensor var_7383_end_mask_0 = const()[name = tensor("op_7383_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7383_cast_fp16 = slice_by_index(begin = var_7383_begin_0, end = var_7383_end_0, end_mask = var_7383_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7383_cast_fp16")]; tensor var_7387_begin_0 = const()[name = tensor("op_7387_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_7387_end_0 = const()[name = tensor("op_7387_end_0"), val = tensor([2, 1152, 1, 77])]; tensor var_7387_end_mask_0 = const()[name = tensor("op_7387_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7387_cast_fp16 = slice_by_index(begin = var_7387_begin_0, end = var_7387_end_0, end_mask = var_7387_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7387_cast_fp16")]; tensor var_7391_begin_0 = const()[name = tensor("op_7391_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_7391_end_0 = const()[name = tensor("op_7391_end_0"), val = tensor([2, 1216, 1, 77])]; tensor var_7391_end_mask_0 = const()[name = tensor("op_7391_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7391_cast_fp16 = slice_by_index(begin = var_7391_begin_0, end = var_7391_end_0, end_mask = var_7391_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7391_cast_fp16")]; tensor var_7395_begin_0 = const()[name = tensor("op_7395_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_7395_end_0 = const()[name = tensor("op_7395_end_0"), val = tensor([2, 1280, 1, 77])]; tensor var_7395_end_mask_0 = const()[name = tensor("op_7395_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7395_cast_fp16 = slice_by_index(begin = var_7395_begin_0, end = var_7395_end_0, end_mask = var_7395_end_mask_0, x = v_35_cast_fp16)[name = tensor("op_7395_cast_fp16")]; tensor var_7399_equation_0 = const()[name = tensor("op_7399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7399_cast_fp16 = einsum(equation = var_7399_equation_0, values = (var_7241_cast_fp16, var_7158_cast_fp16))[name = tensor("op_7399_cast_fp16")]; tensor var_7400_to_fp16 = const()[name = tensor("op_7400_to_fp16"), val = tensor(0x1p-3)]; tensor aw_481_cast_fp16 = mul(x = var_7399_cast_fp16, y = var_7400_to_fp16)[name = tensor("aw_481_cast_fp16")]; tensor var_7403_equation_0 = const()[name = tensor("op_7403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7403_cast_fp16 = einsum(equation = var_7403_equation_0, values = (var_7245_cast_fp16, var_7162_cast_fp16))[name = tensor("op_7403_cast_fp16")]; tensor var_7404_to_fp16 = const()[name = tensor("op_7404_to_fp16"), val = tensor(0x1p-3)]; tensor aw_483_cast_fp16 = mul(x = var_7403_cast_fp16, y = var_7404_to_fp16)[name = tensor("aw_483_cast_fp16")]; tensor var_7407_equation_0 = const()[name = tensor("op_7407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7407_cast_fp16 = einsum(equation = var_7407_equation_0, values = (var_7249_cast_fp16, var_7166_cast_fp16))[name = tensor("op_7407_cast_fp16")]; tensor var_7408_to_fp16 = const()[name = tensor("op_7408_to_fp16"), val = tensor(0x1p-3)]; tensor aw_485_cast_fp16 = mul(x = var_7407_cast_fp16, y = var_7408_to_fp16)[name = tensor("aw_485_cast_fp16")]; tensor var_7411_equation_0 = const()[name = tensor("op_7411_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7411_cast_fp16 = einsum(equation = var_7411_equation_0, values = (var_7253_cast_fp16, var_7170_cast_fp16))[name = tensor("op_7411_cast_fp16")]; tensor var_7412_to_fp16 = const()[name = tensor("op_7412_to_fp16"), val = tensor(0x1p-3)]; tensor aw_487_cast_fp16 = mul(x = var_7411_cast_fp16, y = var_7412_to_fp16)[name = tensor("aw_487_cast_fp16")]; tensor var_7415_equation_0 = const()[name = tensor("op_7415_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7415_cast_fp16 = einsum(equation = var_7415_equation_0, values = (var_7257_cast_fp16, var_7174_cast_fp16))[name = tensor("op_7415_cast_fp16")]; tensor var_7416_to_fp16 = const()[name = tensor("op_7416_to_fp16"), val = tensor(0x1p-3)]; tensor aw_489_cast_fp16 = mul(x = var_7415_cast_fp16, y = var_7416_to_fp16)[name = tensor("aw_489_cast_fp16")]; tensor var_7419_equation_0 = const()[name = tensor("op_7419_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7419_cast_fp16 = einsum(equation = var_7419_equation_0, values = (var_7261_cast_fp16, var_7178_cast_fp16))[name = tensor("op_7419_cast_fp16")]; tensor var_7420_to_fp16 = const()[name = tensor("op_7420_to_fp16"), val = tensor(0x1p-3)]; tensor aw_491_cast_fp16 = mul(x = var_7419_cast_fp16, y = var_7420_to_fp16)[name = tensor("aw_491_cast_fp16")]; tensor var_7423_equation_0 = const()[name = tensor("op_7423_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7423_cast_fp16 = einsum(equation = var_7423_equation_0, values = (var_7265_cast_fp16, var_7182_cast_fp16))[name = tensor("op_7423_cast_fp16")]; tensor var_7424_to_fp16 = const()[name = tensor("op_7424_to_fp16"), val = tensor(0x1p-3)]; tensor aw_493_cast_fp16 = mul(x = var_7423_cast_fp16, y = var_7424_to_fp16)[name = tensor("aw_493_cast_fp16")]; tensor var_7427_equation_0 = const()[name = tensor("op_7427_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7427_cast_fp16 = einsum(equation = var_7427_equation_0, values = (var_7269_cast_fp16, var_7186_cast_fp16))[name = tensor("op_7427_cast_fp16")]; tensor var_7428_to_fp16 = const()[name = tensor("op_7428_to_fp16"), val = tensor(0x1p-3)]; tensor aw_495_cast_fp16 = mul(x = var_7427_cast_fp16, y = var_7428_to_fp16)[name = tensor("aw_495_cast_fp16")]; tensor var_7431_equation_0 = const()[name = tensor("op_7431_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7431_cast_fp16 = einsum(equation = var_7431_equation_0, values = (var_7273_cast_fp16, var_7190_cast_fp16))[name = tensor("op_7431_cast_fp16")]; tensor var_7432_to_fp16 = const()[name = tensor("op_7432_to_fp16"), val = tensor(0x1p-3)]; tensor aw_497_cast_fp16 = mul(x = var_7431_cast_fp16, y = var_7432_to_fp16)[name = tensor("aw_497_cast_fp16")]; tensor var_7435_equation_0 = const()[name = tensor("op_7435_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7435_cast_fp16 = einsum(equation = var_7435_equation_0, values = (var_7277_cast_fp16, var_7194_cast_fp16))[name = tensor("op_7435_cast_fp16")]; tensor var_7436_to_fp16 = const()[name = tensor("op_7436_to_fp16"), val = tensor(0x1p-3)]; tensor aw_499_cast_fp16 = mul(x = var_7435_cast_fp16, y = var_7436_to_fp16)[name = tensor("aw_499_cast_fp16")]; tensor var_7439_equation_0 = const()[name = tensor("op_7439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7439_cast_fp16 = einsum(equation = var_7439_equation_0, values = (var_7281_cast_fp16, var_7198_cast_fp16))[name = tensor("op_7439_cast_fp16")]; tensor var_7440_to_fp16 = const()[name = tensor("op_7440_to_fp16"), val = tensor(0x1p-3)]; tensor aw_501_cast_fp16 = mul(x = var_7439_cast_fp16, y = var_7440_to_fp16)[name = tensor("aw_501_cast_fp16")]; tensor var_7443_equation_0 = const()[name = tensor("op_7443_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7443_cast_fp16 = einsum(equation = var_7443_equation_0, values = (var_7285_cast_fp16, var_7202_cast_fp16))[name = tensor("op_7443_cast_fp16")]; tensor var_7444_to_fp16 = const()[name = tensor("op_7444_to_fp16"), val = tensor(0x1p-3)]; tensor aw_503_cast_fp16 = mul(x = var_7443_cast_fp16, y = var_7444_to_fp16)[name = tensor("aw_503_cast_fp16")]; tensor var_7447_equation_0 = const()[name = tensor("op_7447_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7447_cast_fp16 = einsum(equation = var_7447_equation_0, values = (var_7289_cast_fp16, var_7206_cast_fp16))[name = tensor("op_7447_cast_fp16")]; tensor var_7448_to_fp16 = const()[name = tensor("op_7448_to_fp16"), val = tensor(0x1p-3)]; tensor aw_505_cast_fp16 = mul(x = var_7447_cast_fp16, y = var_7448_to_fp16)[name = tensor("aw_505_cast_fp16")]; tensor var_7451_equation_0 = const()[name = tensor("op_7451_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7451_cast_fp16 = einsum(equation = var_7451_equation_0, values = (var_7293_cast_fp16, var_7210_cast_fp16))[name = tensor("op_7451_cast_fp16")]; tensor var_7452_to_fp16 = const()[name = tensor("op_7452_to_fp16"), val = tensor(0x1p-3)]; tensor aw_507_cast_fp16 = mul(x = var_7451_cast_fp16, y = var_7452_to_fp16)[name = tensor("aw_507_cast_fp16")]; tensor var_7455_equation_0 = const()[name = tensor("op_7455_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7455_cast_fp16 = einsum(equation = var_7455_equation_0, values = (var_7297_cast_fp16, var_7214_cast_fp16))[name = tensor("op_7455_cast_fp16")]; tensor var_7456_to_fp16 = const()[name = tensor("op_7456_to_fp16"), val = tensor(0x1p-3)]; tensor aw_509_cast_fp16 = mul(x = var_7455_cast_fp16, y = var_7456_to_fp16)[name = tensor("aw_509_cast_fp16")]; tensor var_7459_equation_0 = const()[name = tensor("op_7459_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7459_cast_fp16 = einsum(equation = var_7459_equation_0, values = (var_7301_cast_fp16, var_7218_cast_fp16))[name = tensor("op_7459_cast_fp16")]; tensor var_7460_to_fp16 = const()[name = tensor("op_7460_to_fp16"), val = tensor(0x1p-3)]; tensor aw_511_cast_fp16 = mul(x = var_7459_cast_fp16, y = var_7460_to_fp16)[name = tensor("aw_511_cast_fp16")]; tensor var_7463_equation_0 = const()[name = tensor("op_7463_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7463_cast_fp16 = einsum(equation = var_7463_equation_0, values = (var_7305_cast_fp16, var_7222_cast_fp16))[name = tensor("op_7463_cast_fp16")]; tensor var_7464_to_fp16 = const()[name = tensor("op_7464_to_fp16"), val = tensor(0x1p-3)]; tensor aw_513_cast_fp16 = mul(x = var_7463_cast_fp16, y = var_7464_to_fp16)[name = tensor("aw_513_cast_fp16")]; tensor var_7467_equation_0 = const()[name = tensor("op_7467_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7467_cast_fp16 = einsum(equation = var_7467_equation_0, values = (var_7309_cast_fp16, var_7226_cast_fp16))[name = tensor("op_7467_cast_fp16")]; tensor var_7468_to_fp16 = const()[name = tensor("op_7468_to_fp16"), val = tensor(0x1p-3)]; tensor aw_515_cast_fp16 = mul(x = var_7467_cast_fp16, y = var_7468_to_fp16)[name = tensor("aw_515_cast_fp16")]; tensor var_7471_equation_0 = const()[name = tensor("op_7471_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7471_cast_fp16 = einsum(equation = var_7471_equation_0, values = (var_7313_cast_fp16, var_7230_cast_fp16))[name = tensor("op_7471_cast_fp16")]; tensor var_7472_to_fp16 = const()[name = tensor("op_7472_to_fp16"), val = tensor(0x1p-3)]; tensor aw_517_cast_fp16 = mul(x = var_7471_cast_fp16, y = var_7472_to_fp16)[name = tensor("aw_517_cast_fp16")]; tensor var_7475_equation_0 = const()[name = tensor("op_7475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7475_cast_fp16 = einsum(equation = var_7475_equation_0, values = (var_7317_cast_fp16, var_7234_cast_fp16))[name = tensor("op_7475_cast_fp16")]; tensor var_7476_to_fp16 = const()[name = tensor("op_7476_to_fp16"), val = tensor(0x1p-3)]; tensor aw_519_cast_fp16 = mul(x = var_7475_cast_fp16, y = var_7476_to_fp16)[name = tensor("aw_519_cast_fp16")]; tensor var_7478_cast_fp16 = softmax(axis = var_5599, x = aw_481_cast_fp16)[name = tensor("op_7478_cast_fp16")]; tensor var_7479_cast_fp16 = softmax(axis = var_5599, x = aw_483_cast_fp16)[name = tensor("op_7479_cast_fp16")]; tensor var_7480_cast_fp16 = softmax(axis = var_5599, x = aw_485_cast_fp16)[name = tensor("op_7480_cast_fp16")]; tensor var_7481_cast_fp16 = softmax(axis = var_5599, x = aw_487_cast_fp16)[name = tensor("op_7481_cast_fp16")]; tensor var_7482_cast_fp16 = softmax(axis = var_5599, x = aw_489_cast_fp16)[name = tensor("op_7482_cast_fp16")]; tensor var_7483_cast_fp16 = softmax(axis = var_5599, x = aw_491_cast_fp16)[name = tensor("op_7483_cast_fp16")]; tensor var_7484_cast_fp16 = softmax(axis = var_5599, x = aw_493_cast_fp16)[name = tensor("op_7484_cast_fp16")]; tensor var_7485_cast_fp16 = softmax(axis = var_5599, x = aw_495_cast_fp16)[name = tensor("op_7485_cast_fp16")]; tensor var_7486_cast_fp16 = softmax(axis = var_5599, x = aw_497_cast_fp16)[name = tensor("op_7486_cast_fp16")]; tensor var_7487_cast_fp16 = softmax(axis = var_5599, x = aw_499_cast_fp16)[name = tensor("op_7487_cast_fp16")]; tensor var_7488_cast_fp16 = softmax(axis = var_5599, x = aw_501_cast_fp16)[name = tensor("op_7488_cast_fp16")]; tensor var_7489_cast_fp16 = softmax(axis = var_5599, x = aw_503_cast_fp16)[name = tensor("op_7489_cast_fp16")]; tensor var_7490_cast_fp16 = softmax(axis = var_5599, x = aw_505_cast_fp16)[name = tensor("op_7490_cast_fp16")]; tensor var_7491_cast_fp16 = softmax(axis = var_5599, x = aw_507_cast_fp16)[name = tensor("op_7491_cast_fp16")]; tensor var_7492_cast_fp16 = softmax(axis = var_5599, x = aw_509_cast_fp16)[name = tensor("op_7492_cast_fp16")]; tensor var_7493_cast_fp16 = softmax(axis = var_5599, x = aw_511_cast_fp16)[name = tensor("op_7493_cast_fp16")]; tensor var_7494_cast_fp16 = softmax(axis = var_5599, x = aw_513_cast_fp16)[name = tensor("op_7494_cast_fp16")]; tensor var_7495_cast_fp16 = softmax(axis = var_5599, x = aw_515_cast_fp16)[name = tensor("op_7495_cast_fp16")]; tensor var_7496_cast_fp16 = softmax(axis = var_5599, x = aw_517_cast_fp16)[name = tensor("op_7496_cast_fp16")]; tensor var_7497_cast_fp16 = softmax(axis = var_5599, x = aw_519_cast_fp16)[name = tensor("op_7497_cast_fp16")]; tensor var_7499_equation_0 = const()[name = tensor("op_7499_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7499_cast_fp16 = einsum(equation = var_7499_equation_0, values = (var_7319_cast_fp16, var_7478_cast_fp16))[name = tensor("op_7499_cast_fp16")]; tensor var_7501_equation_0 = const()[name = tensor("op_7501_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7501_cast_fp16 = einsum(equation = var_7501_equation_0, values = (var_7323_cast_fp16, var_7479_cast_fp16))[name = tensor("op_7501_cast_fp16")]; tensor var_7503_equation_0 = const()[name = tensor("op_7503_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7503_cast_fp16 = einsum(equation = var_7503_equation_0, values = (var_7327_cast_fp16, var_7480_cast_fp16))[name = tensor("op_7503_cast_fp16")]; tensor var_7505_equation_0 = const()[name = tensor("op_7505_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7505_cast_fp16 = einsum(equation = var_7505_equation_0, values = (var_7331_cast_fp16, var_7481_cast_fp16))[name = tensor("op_7505_cast_fp16")]; tensor var_7507_equation_0 = const()[name = tensor("op_7507_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7507_cast_fp16 = einsum(equation = var_7507_equation_0, values = (var_7335_cast_fp16, var_7482_cast_fp16))[name = tensor("op_7507_cast_fp16")]; tensor var_7509_equation_0 = const()[name = tensor("op_7509_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7509_cast_fp16 = einsum(equation = var_7509_equation_0, values = (var_7339_cast_fp16, var_7483_cast_fp16))[name = tensor("op_7509_cast_fp16")]; tensor var_7511_equation_0 = const()[name = tensor("op_7511_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7511_cast_fp16 = einsum(equation = var_7511_equation_0, values = (var_7343_cast_fp16, var_7484_cast_fp16))[name = tensor("op_7511_cast_fp16")]; tensor var_7513_equation_0 = const()[name = tensor("op_7513_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7513_cast_fp16 = einsum(equation = var_7513_equation_0, values = (var_7347_cast_fp16, var_7485_cast_fp16))[name = tensor("op_7513_cast_fp16")]; tensor var_7515_equation_0 = const()[name = tensor("op_7515_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7515_cast_fp16 = einsum(equation = var_7515_equation_0, values = (var_7351_cast_fp16, var_7486_cast_fp16))[name = tensor("op_7515_cast_fp16")]; tensor var_7517_equation_0 = const()[name = tensor("op_7517_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7517_cast_fp16 = einsum(equation = var_7517_equation_0, values = (var_7355_cast_fp16, var_7487_cast_fp16))[name = tensor("op_7517_cast_fp16")]; tensor var_7519_equation_0 = const()[name = tensor("op_7519_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7519_cast_fp16 = einsum(equation = var_7519_equation_0, values = (var_7359_cast_fp16, var_7488_cast_fp16))[name = tensor("op_7519_cast_fp16")]; tensor var_7521_equation_0 = const()[name = tensor("op_7521_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7521_cast_fp16 = einsum(equation = var_7521_equation_0, values = (var_7363_cast_fp16, var_7489_cast_fp16))[name = tensor("op_7521_cast_fp16")]; tensor var_7523_equation_0 = const()[name = tensor("op_7523_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7523_cast_fp16 = einsum(equation = var_7523_equation_0, values = (var_7367_cast_fp16, var_7490_cast_fp16))[name = tensor("op_7523_cast_fp16")]; tensor var_7525_equation_0 = const()[name = tensor("op_7525_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7525_cast_fp16 = einsum(equation = var_7525_equation_0, values = (var_7371_cast_fp16, var_7491_cast_fp16))[name = tensor("op_7525_cast_fp16")]; tensor var_7527_equation_0 = const()[name = tensor("op_7527_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7527_cast_fp16 = einsum(equation = var_7527_equation_0, values = (var_7375_cast_fp16, var_7492_cast_fp16))[name = tensor("op_7527_cast_fp16")]; tensor var_7529_equation_0 = const()[name = tensor("op_7529_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7529_cast_fp16 = einsum(equation = var_7529_equation_0, values = (var_7379_cast_fp16, var_7493_cast_fp16))[name = tensor("op_7529_cast_fp16")]; tensor var_7531_equation_0 = const()[name = tensor("op_7531_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7531_cast_fp16 = einsum(equation = var_7531_equation_0, values = (var_7383_cast_fp16, var_7494_cast_fp16))[name = tensor("op_7531_cast_fp16")]; tensor var_7533_equation_0 = const()[name = tensor("op_7533_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7533_cast_fp16 = einsum(equation = var_7533_equation_0, values = (var_7387_cast_fp16, var_7495_cast_fp16))[name = tensor("op_7533_cast_fp16")]; tensor var_7535_equation_0 = const()[name = tensor("op_7535_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7535_cast_fp16 = einsum(equation = var_7535_equation_0, values = (var_7391_cast_fp16, var_7496_cast_fp16))[name = tensor("op_7535_cast_fp16")]; tensor var_7537_equation_0 = const()[name = tensor("op_7537_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7537_cast_fp16 = einsum(equation = var_7537_equation_0, values = (var_7395_cast_fp16, var_7497_cast_fp16))[name = tensor("op_7537_cast_fp16")]; tensor input_329_interleave_0 = const()[name = tensor("input_329_interleave_0"), val = tensor(false)]; tensor input_329_cast_fp16 = concat(axis = var_5599, interleave = input_329_interleave_0, values = (var_7499_cast_fp16, var_7501_cast_fp16, var_7503_cast_fp16, var_7505_cast_fp16, var_7507_cast_fp16, var_7509_cast_fp16, var_7511_cast_fp16, var_7513_cast_fp16, var_7515_cast_fp16, var_7517_cast_fp16, var_7519_cast_fp16, var_7521_cast_fp16, var_7523_cast_fp16, var_7525_cast_fp16, var_7527_cast_fp16, var_7529_cast_fp16, var_7531_cast_fp16, var_7533_cast_fp16, var_7535_cast_fp16, var_7537_cast_fp16))[name = tensor("input_329_cast_fp16")]; tensor var_7543 = const()[name = tensor("op_7543"), val = tensor([1, 1])]; tensor var_7545 = const()[name = tensor("op_7545"), val = tensor([1, 1])]; tensor var_7547_pad_type_0 = const()[name = tensor("op_7547_pad_type_0"), val = tensor("custom")]; tensor var_7547_pad_0 = const()[name = tensor("op_7547_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(426820544)))]; tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430097408)))]; tensor var_7547_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_7545, groups = var_5599, pad = var_7547_pad_0, pad_type = var_7547_pad_type_0, strides = var_7543, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_329_cast_fp16)[name = tensor("op_7547_cast_fp16")]; tensor inputs_53_cast_fp16 = add(x = var_7547_cast_fp16, y = inputs_51_cast_fp16)[name = tensor("inputs_53_cast_fp16")]; tensor var_7551 = const()[name = tensor("op_7551"), val = tensor([1])]; tensor channels_mean_53_cast_fp16 = reduce_mean(axes = var_7551, keep_dims = var_5594, x = inputs_53_cast_fp16)[name = tensor("channels_mean_53_cast_fp16")]; tensor zero_mean_53_cast_fp16 = sub(x = inputs_53_cast_fp16, y = channels_mean_53_cast_fp16)[name = tensor("zero_mean_53_cast_fp16")]; tensor zero_mean_sq_53_cast_fp16 = mul(x = zero_mean_53_cast_fp16, y = zero_mean_53_cast_fp16)[name = tensor("zero_mean_sq_53_cast_fp16")]; tensor var_7555 = const()[name = tensor("op_7555"), val = tensor([1])]; tensor var_7556_cast_fp16 = reduce_mean(axes = var_7555, keep_dims = var_5594, x = zero_mean_sq_53_cast_fp16)[name = tensor("op_7556_cast_fp16")]; tensor var_7557_to_fp16 = const()[name = tensor("op_7557_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_7558_cast_fp16 = add(x = var_7556_cast_fp16, y = var_7557_to_fp16)[name = tensor("op_7558_cast_fp16")]; tensor denom_53_epsilon_0_to_fp16 = const()[name = tensor("denom_53_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_53_cast_fp16 = rsqrt(epsilon = denom_53_epsilon_0_to_fp16, x = var_7558_cast_fp16)[name = tensor("denom_53_cast_fp16")]; tensor out_53_cast_fp16 = mul(x = zero_mean_53_cast_fp16, y = denom_53_cast_fp16)[name = tensor("out_53_cast_fp16")]; tensor var_7562_to_fp16 = const()[name = tensor("op_7562_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430100032)))]; tensor var_7563_cast_fp16 = add(x = out_53_cast_fp16, y = var_7562_to_fp16)[name = tensor("op_7563_cast_fp16")]; tensor var_7565_to_fp16 = const()[name = tensor("op_7565_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430102656)))]; tensor input_331_cast_fp16 = mul(x = var_7563_cast_fp16, y = var_7565_to_fp16)[name = tensor("input_331_cast_fp16")]; tensor var_7573 = const()[name = tensor("op_7573"), val = tensor([1, 1])]; tensor var_7575 = const()[name = tensor("op_7575"), val = tensor([1, 1])]; tensor var_7577_pad_type_0 = const()[name = tensor("op_7577_pad_type_0"), val = tensor("custom")]; tensor var_7577_pad_0 = const()[name = tensor("op_7577_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430105280)))]; tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(456319744)))]; tensor var_7577_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_7575, groups = var_5599, pad = var_7577_pad_0, pad_type = var_7577_pad_type_0, strides = var_7573, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_331_cast_fp16)[name = tensor("op_7577_cast_fp16")]; tensor var_7578_split_sizes_0 = const()[name = tensor("op_7578_split_sizes_0"), val = tensor([5120, 5120])]; tensor var_7578_axis_0 = const()[name = tensor("op_7578_axis_0"), val = tensor(1)]; tensor var_7578_cast_fp16_0, tensor var_7578_cast_fp16_1 = split(axis = var_7578_axis_0, split_sizes = var_7578_split_sizes_0, x = var_7577_cast_fp16)[name = tensor("op_7578_cast_fp16")]; tensor var_7580_mode_0 = const()[name = tensor("op_7580_mode_0"), val = tensor("EXACT")]; tensor var_7580_cast_fp16 = gelu(mode = var_7580_mode_0, x = var_7578_cast_fp16_1)[name = tensor("op_7580_cast_fp16")]; tensor input_333_cast_fp16 = mul(x = var_7578_cast_fp16_0, y = var_7580_cast_fp16)[name = tensor("input_333_cast_fp16")]; tensor var_7584 = const()[name = tensor("op_7584"), val = tensor([1, 1])]; tensor var_7586 = const()[name = tensor("op_7586"), val = tensor([1, 1])]; tensor var_7588_pad_type_0 = const()[name = tensor("op_7588_pad_type_0"), val = tensor("custom")]; tensor var_7588_pad_0 = const()[name = tensor("op_7588_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(456340288)))]; tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(469447552)))]; tensor var_7588_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_7586, groups = var_5599, pad = var_7588_pad_0, pad_type = var_7588_pad_type_0, strides = var_7584, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_333_cast_fp16)[name = tensor("op_7588_cast_fp16")]; tensor hidden_states_193_cast_fp16 = add(x = var_7588_cast_fp16, y = inputs_53_cast_fp16)[name = tensor("hidden_states_193_cast_fp16")]; tensor var_7590 = const()[name = tensor("op_7590"), val = tensor([2, 1280, 12, 12])]; tensor input_335_cast_fp16 = reshape(shape = var_7590, x = hidden_states_193_cast_fp16)[name = tensor("input_335_cast_fp16")]; tensor var_7594 = const()[name = tensor("op_7594"), val = tensor([1, 1])]; tensor var_7596 = const()[name = tensor("op_7596"), val = tensor([1, 1])]; tensor hidden_states_195_pad_type_0 = const()[name = tensor("hidden_states_195_pad_type_0"), val = tensor("custom")]; tensor hidden_states_195_pad_0 = const()[name = tensor("hidden_states_195_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(469450176)))]; tensor up_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472727040)))]; tensor hidden_states_195_cast_fp16 = conv(bias = up_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_7596, groups = var_5599, pad = hidden_states_195_pad_0, pad_type = hidden_states_195_pad_type_0, strides = var_7594, weight = up_blocks_1_attentions_1_proj_out_weight_to_fp16, x = input_335_cast_fp16)[name = tensor("hidden_states_195_cast_fp16")]; tensor hidden_states_197_cast_fp16 = add(x = hidden_states_195_cast_fp16, y = hidden_states_183_cast_fp16)[name = tensor("hidden_states_197_cast_fp16")]; tensor input_337_interleave_0 = const()[name = tensor("input_337_interleave_0"), val = tensor(false)]; tensor cast_6 = cast(dtype = cast_8_dtype_0, x = input_117_cast_fp16)[name = tensor("cast_6")]; tensor input_337_cast_fp16 = concat(axis = var_5599, interleave = input_337_interleave_0, values = (hidden_states_197_cast_fp16, cast_6))[name = tensor("input_337_cast_fp16")]; tensor reshape_156_shape_0 = const()[name = tensor("reshape_156_shape_0"), val = tensor([2, 32, 60, 12, 12])]; tensor reshape_156_cast_fp16 = reshape(shape = reshape_156_shape_0, x = input_337_cast_fp16)[name = tensor("reshape_156_cast_fp16")]; tensor reduce_mean_117_axes_0 = const()[name = tensor("reduce_mean_117_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_117_keep_dims_0 = const()[name = tensor("reduce_mean_117_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_117_cast_fp16 = reduce_mean(axes = reduce_mean_117_axes_0, keep_dims = reduce_mean_117_keep_dims_0, x = reshape_156_cast_fp16)[name = tensor("reduce_mean_117_cast_fp16")]; tensor sub_78_cast_fp16 = sub(x = reshape_156_cast_fp16, y = reduce_mean_117_cast_fp16)[name = tensor("sub_78_cast_fp16")]; tensor square_39_cast_fp16 = square(x = sub_78_cast_fp16)[name = tensor("square_39_cast_fp16")]; tensor reduce_mean_119_axes_0 = const()[name = tensor("reduce_mean_119_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_119_keep_dims_0 = const()[name = tensor("reduce_mean_119_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_119_cast_fp16 = reduce_mean(axes = reduce_mean_119_axes_0, keep_dims = reduce_mean_119_keep_dims_0, x = square_39_cast_fp16)[name = tensor("reduce_mean_119_cast_fp16")]; tensor add_78_y_0_to_fp16 = const()[name = tensor("add_78_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_78_cast_fp16 = add(x = reduce_mean_119_cast_fp16, y = add_78_y_0_to_fp16)[name = tensor("add_78_cast_fp16")]; tensor sqrt_39_cast_fp16 = sqrt(x = add_78_cast_fp16)[name = tensor("sqrt_39_cast_fp16")]; tensor real_div_39_cast_fp16 = real_div(x = sub_78_cast_fp16, y = sqrt_39_cast_fp16)[name = tensor("real_div_39_cast_fp16")]; tensor reshape_157_shape_0 = const()[name = tensor("reshape_157_shape_0"), val = tensor([2, 1920, 12, 12])]; tensor reshape_157_cast_fp16 = reshape(shape = reshape_157_shape_0, x = real_div_39_cast_fp16)[name = tensor("reshape_157_cast_fp16")]; tensor add_79_mean_0_to_fp16 = const()[name = tensor("add_79_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472729664)))]; tensor add_79_variance_0_to_fp16 = const()[name = tensor("add_79_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472733568)))]; tensor add_79_gamma_0_to_fp16 = const()[name = tensor("add_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472737472)))]; tensor add_79_beta_0_to_fp16 = const()[name = tensor("add_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472741376)))]; tensor add_79_epsilon_0_to_fp16 = const()[name = tensor("add_79_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_79_cast_fp16 = batch_norm(beta = add_79_beta_0_to_fp16, epsilon = add_79_epsilon_0_to_fp16, gamma = add_79_gamma_0_to_fp16, mean = add_79_mean_0_to_fp16, variance = add_79_variance_0_to_fp16, x = reshape_157_cast_fp16)[name = tensor("add_79_cast_fp16")]; tensor input_341_cast_fp16 = silu(x = add_79_cast_fp16)[name = tensor("input_341_cast_fp16")]; tensor var_7614 = const()[name = tensor("op_7614"), val = tensor([1, 1])]; tensor var_7616 = const()[name = tensor("op_7616"), val = tensor([1, 1])]; tensor hidden_states_199_pad_type_0 = const()[name = tensor("hidden_states_199_pad_type_0"), val = tensor("custom")]; tensor hidden_states_199_pad_0 = const()[name = tensor("hidden_states_199_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_1_resnets_2_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472745280)))]; tensor up_blocks_1_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(516982144)))]; tensor hidden_states_199_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv1_bias_to_fp16, dilations = var_7616, groups = var_5599, pad = hidden_states_199_pad_0, pad_type = hidden_states_199_pad_type_0, strides = var_7614, weight = up_blocks_1_resnets_2_conv1_weight_to_fp16, x = input_341_cast_fp16)[name = tensor("hidden_states_199_cast_fp16")]; tensor var_7622 = const()[name = tensor("op_7622"), val = tensor([1, 1])]; tensor var_7624 = const()[name = tensor("op_7624"), val = tensor([1, 1])]; tensor temb_31_pad_type_0 = const()[name = tensor("temb_31_pad_type_0"), val = tensor("custom")]; tensor temb_31_pad_0 = const()[name = tensor("temb_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(516984768)))]; tensor up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520261632)))]; tensor temb_31_cast_fp16 = conv(bias = up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_7624, groups = var_5599, pad = temb_31_pad_0, pad_type = temb_31_pad_type_0, strides = var_7622, weight = up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16, x = cast_12)[name = tensor("temb_31_cast_fp16")]; tensor input_345_cast_fp16 = add(x = hidden_states_199_cast_fp16, y = temb_31_cast_fp16)[name = tensor("input_345_cast_fp16")]; tensor reshape_160_shape_0 = const()[name = tensor("reshape_160_shape_0"), val = tensor([2, 32, 40, 12, 12])]; tensor reshape_160_cast_fp16 = reshape(shape = reshape_160_shape_0, x = input_345_cast_fp16)[name = tensor("reshape_160_cast_fp16")]; tensor reduce_mean_120_axes_0 = const()[name = tensor("reduce_mean_120_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_120_keep_dims_0 = const()[name = tensor("reduce_mean_120_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_120_cast_fp16 = reduce_mean(axes = reduce_mean_120_axes_0, keep_dims = reduce_mean_120_keep_dims_0, x = reshape_160_cast_fp16)[name = tensor("reduce_mean_120_cast_fp16")]; tensor sub_80_cast_fp16 = sub(x = reshape_160_cast_fp16, y = reduce_mean_120_cast_fp16)[name = tensor("sub_80_cast_fp16")]; tensor square_40_cast_fp16 = square(x = sub_80_cast_fp16)[name = tensor("square_40_cast_fp16")]; tensor reduce_mean_122_axes_0 = const()[name = tensor("reduce_mean_122_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_122_keep_dims_0 = const()[name = tensor("reduce_mean_122_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_122_cast_fp16 = reduce_mean(axes = reduce_mean_122_axes_0, keep_dims = reduce_mean_122_keep_dims_0, x = square_40_cast_fp16)[name = tensor("reduce_mean_122_cast_fp16")]; tensor add_80_y_0_to_fp16 = const()[name = tensor("add_80_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_80_cast_fp16 = add(x = reduce_mean_122_cast_fp16, y = add_80_y_0_to_fp16)[name = tensor("add_80_cast_fp16")]; tensor sqrt_40_cast_fp16 = sqrt(x = add_80_cast_fp16)[name = tensor("sqrt_40_cast_fp16")]; tensor real_div_40_cast_fp16 = real_div(x = sub_80_cast_fp16, y = sqrt_40_cast_fp16)[name = tensor("real_div_40_cast_fp16")]; tensor reshape_161_shape_0 = const()[name = tensor("reshape_161_shape_0"), val = tensor([2, 1280, 12, 12])]; tensor reshape_161_cast_fp16 = reshape(shape = reshape_161_shape_0, x = real_div_40_cast_fp16)[name = tensor("reshape_161_cast_fp16")]; tensor add_81_gamma_0_to_fp16 = const()[name = tensor("add_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520264256)))]; tensor add_81_beta_0_to_fp16 = const()[name = tensor("add_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520266880)))]; tensor add_81_epsilon_0_to_fp16 = const()[name = tensor("add_81_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_81_cast_fp16 = batch_norm(beta = add_81_beta_0_to_fp16, epsilon = add_81_epsilon_0_to_fp16, gamma = add_81_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_161_cast_fp16)[name = tensor("add_81_cast_fp16")]; tensor input_349_cast_fp16 = silu(x = add_81_cast_fp16)[name = tensor("input_349_cast_fp16")]; tensor var_7634 = const()[name = tensor("op_7634"), val = tensor([1, 1])]; tensor var_7636 = const()[name = tensor("op_7636"), val = tensor([1, 1])]; tensor hidden_states_201_pad_type_0 = const()[name = tensor("hidden_states_201_pad_type_0"), val = tensor("custom")]; tensor hidden_states_201_pad_0 = const()[name = tensor("hidden_states_201_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_1_resnets_2_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520269504)))]; tensor up_blocks_1_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549760768)))]; tensor hidden_states_201_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv2_bias_to_fp16, dilations = var_7636, groups = var_5599, pad = hidden_states_201_pad_0, pad_type = hidden_states_201_pad_type_0, strides = var_7634, weight = up_blocks_1_resnets_2_conv2_weight_to_fp16, x = input_349_cast_fp16)[name = tensor("hidden_states_201_cast_fp16")]; tensor var_7641 = const()[name = tensor("op_7641"), val = tensor([1, 1])]; tensor var_7643 = const()[name = tensor("op_7643"), val = tensor([1, 1])]; tensor x_15_pad_type_0 = const()[name = tensor("x_15_pad_type_0"), val = tensor("custom")]; tensor x_15_pad_0 = const()[name = tensor("x_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549763392)))]; tensor up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(554678656)))]; tensor x_15_cast_fp16 = conv(bias = up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_7643, groups = var_5599, pad = x_15_pad_0, pad_type = x_15_pad_type_0, strides = var_7641, weight = up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16, x = input_337_cast_fp16)[name = tensor("x_15_cast_fp16")]; tensor hidden_states_203_cast_fp16 = add(x = x_15_cast_fp16, y = hidden_states_201_cast_fp16)[name = tensor("hidden_states_203_cast_fp16")]; tensor reshape_164_shape_0 = const()[name = tensor("reshape_164_shape_0"), val = tensor([2, 32, 40, 12, 12])]; tensor reshape_164_cast_fp16 = reshape(shape = reshape_164_shape_0, x = hidden_states_203_cast_fp16)[name = tensor("reshape_164_cast_fp16")]; tensor reduce_mean_123_axes_0 = const()[name = tensor("reduce_mean_123_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_123_keep_dims_0 = const()[name = tensor("reduce_mean_123_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_123_cast_fp16 = reduce_mean(axes = reduce_mean_123_axes_0, keep_dims = reduce_mean_123_keep_dims_0, x = reshape_164_cast_fp16)[name = tensor("reduce_mean_123_cast_fp16")]; tensor sub_82_cast_fp16 = sub(x = reshape_164_cast_fp16, y = reduce_mean_123_cast_fp16)[name = tensor("sub_82_cast_fp16")]; tensor square_41_cast_fp16 = square(x = sub_82_cast_fp16)[name = tensor("square_41_cast_fp16")]; tensor reduce_mean_125_axes_0 = const()[name = tensor("reduce_mean_125_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_125_keep_dims_0 = const()[name = tensor("reduce_mean_125_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_125_cast_fp16 = reduce_mean(axes = reduce_mean_125_axes_0, keep_dims = reduce_mean_125_keep_dims_0, x = square_41_cast_fp16)[name = tensor("reduce_mean_125_cast_fp16")]; tensor add_82_y_0_to_fp16 = const()[name = tensor("add_82_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_82_cast_fp16 = add(x = reduce_mean_125_cast_fp16, y = add_82_y_0_to_fp16)[name = tensor("add_82_cast_fp16")]; tensor sqrt_41_cast_fp16 = sqrt(x = add_82_cast_fp16)[name = tensor("sqrt_41_cast_fp16")]; tensor real_div_41_cast_fp16 = real_div(x = sub_82_cast_fp16, y = sqrt_41_cast_fp16)[name = tensor("real_div_41_cast_fp16")]; tensor reshape_165_shape_0 = const()[name = tensor("reshape_165_shape_0"), val = tensor([2, 1280, 12, 12])]; tensor reshape_165_cast_fp16 = reshape(shape = reshape_165_shape_0, x = real_div_41_cast_fp16)[name = tensor("reshape_165_cast_fp16")]; tensor add_83_gamma_0_to_fp16 = const()[name = tensor("add_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(554681280)))]; tensor add_83_beta_0_to_fp16 = const()[name = tensor("add_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(554683904)))]; tensor add_83_epsilon_0_to_fp16 = const()[name = tensor("add_83_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_83_cast_fp16 = batch_norm(beta = add_83_beta_0_to_fp16, epsilon = add_83_epsilon_0_to_fp16, gamma = add_83_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_165_cast_fp16)[name = tensor("add_83_cast_fp16")]; tensor var_7663 = const()[name = tensor("op_7663"), val = tensor([1, 1])]; tensor var_7665 = const()[name = tensor("op_7665"), val = tensor([1, 1])]; tensor hidden_states_205_pad_type_0 = const()[name = tensor("hidden_states_205_pad_type_0"), val = tensor("custom")]; tensor hidden_states_205_pad_0 = const()[name = tensor("hidden_states_205_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(554686528)))]; tensor up_blocks_1_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(557963392)))]; tensor hidden_states_205_cast_fp16 = conv(bias = up_blocks_1_attentions_2_proj_in_bias_to_fp16, dilations = var_7665, groups = var_5599, pad = hidden_states_205_pad_0, pad_type = hidden_states_205_pad_type_0, strides = var_7663, weight = up_blocks_1_attentions_2_proj_in_weight_to_fp16, x = add_83_cast_fp16)[name = tensor("hidden_states_205_cast_fp16")]; tensor var_7670 = const()[name = tensor("op_7670"), val = tensor([2, 1280, 1, 144])]; tensor inputs_55_cast_fp16 = reshape(shape = var_7670, x = hidden_states_205_cast_fp16)[name = tensor("inputs_55_cast_fp16")]; tensor var_7680 = const()[name = tensor("op_7680"), val = tensor([1])]; tensor channels_mean_55_cast_fp16 = reduce_mean(axes = var_7680, keep_dims = var_5594, x = inputs_55_cast_fp16)[name = tensor("channels_mean_55_cast_fp16")]; tensor zero_mean_55_cast_fp16 = sub(x = inputs_55_cast_fp16, y = channels_mean_55_cast_fp16)[name = tensor("zero_mean_55_cast_fp16")]; tensor zero_mean_sq_55_cast_fp16 = mul(x = zero_mean_55_cast_fp16, y = zero_mean_55_cast_fp16)[name = tensor("zero_mean_sq_55_cast_fp16")]; tensor var_7684 = const()[name = tensor("op_7684"), val = tensor([1])]; tensor var_7685_cast_fp16 = reduce_mean(axes = var_7684, keep_dims = var_5594, x = zero_mean_sq_55_cast_fp16)[name = tensor("op_7685_cast_fp16")]; tensor var_7686_to_fp16 = const()[name = tensor("op_7686_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_7687_cast_fp16 = add(x = var_7685_cast_fp16, y = var_7686_to_fp16)[name = tensor("op_7687_cast_fp16")]; tensor denom_55_epsilon_0_to_fp16 = const()[name = tensor("denom_55_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_55_cast_fp16 = rsqrt(epsilon = denom_55_epsilon_0_to_fp16, x = var_7687_cast_fp16)[name = tensor("denom_55_cast_fp16")]; tensor out_55_cast_fp16 = mul(x = zero_mean_55_cast_fp16, y = denom_55_cast_fp16)[name = tensor("out_55_cast_fp16")]; tensor var_7691_to_fp16 = const()[name = tensor("op_7691_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(557966016)))]; tensor var_7692_cast_fp16 = add(x = out_55_cast_fp16, y = var_7691_to_fp16)[name = tensor("op_7692_cast_fp16")]; tensor var_7694_to_fp16 = const()[name = tensor("op_7694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(557968640)))]; tensor hidden_states_207_cast_fp16 = mul(x = var_7692_cast_fp16, y = var_7694_to_fp16)[name = tensor("hidden_states_207_cast_fp16")]; tensor var_7701 = const()[name = tensor("op_7701"), val = tensor([1, 1])]; tensor var_7703 = const()[name = tensor("op_7703"), val = tensor([1, 1])]; tensor q_37_pad_type_0 = const()[name = tensor("q_37_pad_type_0"), val = tensor("custom")]; tensor q_37_pad_0 = const()[name = tensor("q_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(557971264)))]; tensor q_37_cast_fp16 = conv(dilations = var_7703, groups = var_5599, pad = q_37_pad_0, pad_type = q_37_pad_type_0, strides = var_7701, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_207_cast_fp16)[name = tensor("q_37_cast_fp16")]; tensor var_7707 = const()[name = tensor("op_7707"), val = tensor([1, 1])]; tensor var_7709 = const()[name = tensor("op_7709"), val = tensor([1, 1])]; tensor k_73_pad_type_0 = const()[name = tensor("k_73_pad_type_0"), val = tensor("custom")]; tensor k_73_pad_0 = const()[name = tensor("k_73_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(561248128)))]; tensor k_73_cast_fp16 = conv(dilations = var_7709, groups = var_5599, pad = k_73_pad_0, pad_type = k_73_pad_type_0, strides = var_7707, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_207_cast_fp16)[name = tensor("k_73_cast_fp16")]; tensor var_7713 = const()[name = tensor("op_7713"), val = tensor([1, 1])]; tensor var_7715 = const()[name = tensor("op_7715"), val = tensor([1, 1])]; tensor v_37_pad_type_0 = const()[name = tensor("v_37_pad_type_0"), val = tensor("custom")]; tensor v_37_pad_0 = const()[name = tensor("v_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564524992)))]; tensor v_37_cast_fp16 = conv(dilations = var_7715, groups = var_5599, pad = v_37_pad_0, pad_type = v_37_pad_type_0, strides = var_7713, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_207_cast_fp16)[name = tensor("v_37_cast_fp16")]; tensor var_7719_begin_0 = const()[name = tensor("op_7719_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7719_end_0 = const()[name = tensor("op_7719_end_0"), val = tensor([2, 64, 1, 144])]; tensor var_7719_end_mask_0 = const()[name = tensor("op_7719_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7719_cast_fp16 = slice_by_index(begin = var_7719_begin_0, end = var_7719_end_0, end_mask = var_7719_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7719_cast_fp16")]; tensor var_7723_begin_0 = const()[name = tensor("op_7723_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_7723_end_0 = const()[name = tensor("op_7723_end_0"), val = tensor([2, 128, 1, 144])]; tensor var_7723_end_mask_0 = const()[name = tensor("op_7723_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7723_cast_fp16 = slice_by_index(begin = var_7723_begin_0, end = var_7723_end_0, end_mask = var_7723_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7723_cast_fp16")]; tensor var_7727_begin_0 = const()[name = tensor("op_7727_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_7727_end_0 = const()[name = tensor("op_7727_end_0"), val = tensor([2, 192, 1, 144])]; tensor var_7727_end_mask_0 = const()[name = tensor("op_7727_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7727_cast_fp16 = slice_by_index(begin = var_7727_begin_0, end = var_7727_end_0, end_mask = var_7727_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7727_cast_fp16")]; tensor var_7731_begin_0 = const()[name = tensor("op_7731_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_7731_end_0 = const()[name = tensor("op_7731_end_0"), val = tensor([2, 256, 1, 144])]; tensor var_7731_end_mask_0 = const()[name = tensor("op_7731_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7731_cast_fp16 = slice_by_index(begin = var_7731_begin_0, end = var_7731_end_0, end_mask = var_7731_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7731_cast_fp16")]; tensor var_7735_begin_0 = const()[name = tensor("op_7735_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_7735_end_0 = const()[name = tensor("op_7735_end_0"), val = tensor([2, 320, 1, 144])]; tensor var_7735_end_mask_0 = const()[name = tensor("op_7735_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7735_cast_fp16 = slice_by_index(begin = var_7735_begin_0, end = var_7735_end_0, end_mask = var_7735_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7735_cast_fp16")]; tensor var_7739_begin_0 = const()[name = tensor("op_7739_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7739_end_0 = const()[name = tensor("op_7739_end_0"), val = tensor([2, 384, 1, 144])]; tensor var_7739_end_mask_0 = const()[name = tensor("op_7739_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7739_cast_fp16 = slice_by_index(begin = var_7739_begin_0, end = var_7739_end_0, end_mask = var_7739_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7739_cast_fp16")]; tensor var_7743_begin_0 = const()[name = tensor("op_7743_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_7743_end_0 = const()[name = tensor("op_7743_end_0"), val = tensor([2, 448, 1, 144])]; tensor var_7743_end_mask_0 = const()[name = tensor("op_7743_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7743_cast_fp16 = slice_by_index(begin = var_7743_begin_0, end = var_7743_end_0, end_mask = var_7743_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7743_cast_fp16")]; tensor var_7747_begin_0 = const()[name = tensor("op_7747_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_7747_end_0 = const()[name = tensor("op_7747_end_0"), val = tensor([2, 512, 1, 144])]; tensor var_7747_end_mask_0 = const()[name = tensor("op_7747_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7747_cast_fp16 = slice_by_index(begin = var_7747_begin_0, end = var_7747_end_0, end_mask = var_7747_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7747_cast_fp16")]; tensor var_7751_begin_0 = const()[name = tensor("op_7751_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_7751_end_0 = const()[name = tensor("op_7751_end_0"), val = tensor([2, 576, 1, 144])]; tensor var_7751_end_mask_0 = const()[name = tensor("op_7751_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7751_cast_fp16 = slice_by_index(begin = var_7751_begin_0, end = var_7751_end_0, end_mask = var_7751_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7751_cast_fp16")]; tensor var_7755_begin_0 = const()[name = tensor("op_7755_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_7755_end_0 = const()[name = tensor("op_7755_end_0"), val = tensor([2, 640, 1, 144])]; tensor var_7755_end_mask_0 = const()[name = tensor("op_7755_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7755_cast_fp16 = slice_by_index(begin = var_7755_begin_0, end = var_7755_end_0, end_mask = var_7755_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7755_cast_fp16")]; tensor var_7759_begin_0 = const()[name = tensor("op_7759_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7759_end_0 = const()[name = tensor("op_7759_end_0"), val = tensor([2, 704, 1, 144])]; tensor var_7759_end_mask_0 = const()[name = tensor("op_7759_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7759_cast_fp16 = slice_by_index(begin = var_7759_begin_0, end = var_7759_end_0, end_mask = var_7759_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7759_cast_fp16")]; tensor var_7763_begin_0 = const()[name = tensor("op_7763_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_7763_end_0 = const()[name = tensor("op_7763_end_0"), val = tensor([2, 768, 1, 144])]; tensor var_7763_end_mask_0 = const()[name = tensor("op_7763_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7763_cast_fp16 = slice_by_index(begin = var_7763_begin_0, end = var_7763_end_0, end_mask = var_7763_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7763_cast_fp16")]; tensor var_7767_begin_0 = const()[name = tensor("op_7767_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_7767_end_0 = const()[name = tensor("op_7767_end_0"), val = tensor([2, 832, 1, 144])]; tensor var_7767_end_mask_0 = const()[name = tensor("op_7767_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7767_cast_fp16 = slice_by_index(begin = var_7767_begin_0, end = var_7767_end_0, end_mask = var_7767_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7767_cast_fp16")]; tensor var_7771_begin_0 = const()[name = tensor("op_7771_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_7771_end_0 = const()[name = tensor("op_7771_end_0"), val = tensor([2, 896, 1, 144])]; tensor var_7771_end_mask_0 = const()[name = tensor("op_7771_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7771_cast_fp16 = slice_by_index(begin = var_7771_begin_0, end = var_7771_end_0, end_mask = var_7771_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7771_cast_fp16")]; tensor var_7775_begin_0 = const()[name = tensor("op_7775_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_7775_end_0 = const()[name = tensor("op_7775_end_0"), val = tensor([2, 960, 1, 144])]; tensor var_7775_end_mask_0 = const()[name = tensor("op_7775_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7775_cast_fp16 = slice_by_index(begin = var_7775_begin_0, end = var_7775_end_0, end_mask = var_7775_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7775_cast_fp16")]; tensor var_7779_begin_0 = const()[name = tensor("op_7779_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_7779_end_0 = const()[name = tensor("op_7779_end_0"), val = tensor([2, 1024, 1, 144])]; tensor var_7779_end_mask_0 = const()[name = tensor("op_7779_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7779_cast_fp16 = slice_by_index(begin = var_7779_begin_0, end = var_7779_end_0, end_mask = var_7779_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7779_cast_fp16")]; tensor var_7783_begin_0 = const()[name = tensor("op_7783_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_7783_end_0 = const()[name = tensor("op_7783_end_0"), val = tensor([2, 1088, 1, 144])]; tensor var_7783_end_mask_0 = const()[name = tensor("op_7783_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7783_cast_fp16 = slice_by_index(begin = var_7783_begin_0, end = var_7783_end_0, end_mask = var_7783_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7783_cast_fp16")]; tensor var_7787_begin_0 = const()[name = tensor("op_7787_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_7787_end_0 = const()[name = tensor("op_7787_end_0"), val = tensor([2, 1152, 1, 144])]; tensor var_7787_end_mask_0 = const()[name = tensor("op_7787_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7787_cast_fp16 = slice_by_index(begin = var_7787_begin_0, end = var_7787_end_0, end_mask = var_7787_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7787_cast_fp16")]; tensor var_7791_begin_0 = const()[name = tensor("op_7791_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_7791_end_0 = const()[name = tensor("op_7791_end_0"), val = tensor([2, 1216, 1, 144])]; tensor var_7791_end_mask_0 = const()[name = tensor("op_7791_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7791_cast_fp16 = slice_by_index(begin = var_7791_begin_0, end = var_7791_end_0, end_mask = var_7791_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7791_cast_fp16")]; tensor var_7795_begin_0 = const()[name = tensor("op_7795_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_7795_end_0 = const()[name = tensor("op_7795_end_0"), val = tensor([2, 1280, 1, 144])]; tensor var_7795_end_mask_0 = const()[name = tensor("op_7795_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7795_cast_fp16 = slice_by_index(begin = var_7795_begin_0, end = var_7795_end_0, end_mask = var_7795_end_mask_0, x = q_37_cast_fp16)[name = tensor("op_7795_cast_fp16")]; tensor k_75_perm_0 = const()[name = tensor("k_75_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_7802_begin_0 = const()[name = tensor("op_7802_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7802_end_0 = const()[name = tensor("op_7802_end_0"), val = tensor([2, 144, 1, 64])]; tensor var_7802_end_mask_0 = const()[name = tensor("op_7802_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_13 = transpose(perm = k_75_perm_0, x = k_73_cast_fp16)[name = tensor("transpose_13")]; tensor var_7802_cast_fp16 = slice_by_index(begin = var_7802_begin_0, end = var_7802_end_0, end_mask = var_7802_end_mask_0, x = transpose_13)[name = tensor("op_7802_cast_fp16")]; tensor var_7806_begin_0 = const()[name = tensor("op_7806_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_7806_end_0 = const()[name = tensor("op_7806_end_0"), val = tensor([2, 144, 1, 128])]; tensor var_7806_end_mask_0 = const()[name = tensor("op_7806_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7806_cast_fp16 = slice_by_index(begin = var_7806_begin_0, end = var_7806_end_0, end_mask = var_7806_end_mask_0, x = transpose_13)[name = tensor("op_7806_cast_fp16")]; tensor var_7810_begin_0 = const()[name = tensor("op_7810_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_7810_end_0 = const()[name = tensor("op_7810_end_0"), val = tensor([2, 144, 1, 192])]; tensor var_7810_end_mask_0 = const()[name = tensor("op_7810_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7810_cast_fp16 = slice_by_index(begin = var_7810_begin_0, end = var_7810_end_0, end_mask = var_7810_end_mask_0, x = transpose_13)[name = tensor("op_7810_cast_fp16")]; tensor var_7814_begin_0 = const()[name = tensor("op_7814_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_7814_end_0 = const()[name = tensor("op_7814_end_0"), val = tensor([2, 144, 1, 256])]; tensor var_7814_end_mask_0 = const()[name = tensor("op_7814_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7814_cast_fp16 = slice_by_index(begin = var_7814_begin_0, end = var_7814_end_0, end_mask = var_7814_end_mask_0, x = transpose_13)[name = tensor("op_7814_cast_fp16")]; tensor var_7818_begin_0 = const()[name = tensor("op_7818_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_7818_end_0 = const()[name = tensor("op_7818_end_0"), val = tensor([2, 144, 1, 320])]; tensor var_7818_end_mask_0 = const()[name = tensor("op_7818_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7818_cast_fp16 = slice_by_index(begin = var_7818_begin_0, end = var_7818_end_0, end_mask = var_7818_end_mask_0, x = transpose_13)[name = tensor("op_7818_cast_fp16")]; tensor var_7822_begin_0 = const()[name = tensor("op_7822_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_7822_end_0 = const()[name = tensor("op_7822_end_0"), val = tensor([2, 144, 1, 384])]; tensor var_7822_end_mask_0 = const()[name = tensor("op_7822_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7822_cast_fp16 = slice_by_index(begin = var_7822_begin_0, end = var_7822_end_0, end_mask = var_7822_end_mask_0, x = transpose_13)[name = tensor("op_7822_cast_fp16")]; tensor var_7826_begin_0 = const()[name = tensor("op_7826_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_7826_end_0 = const()[name = tensor("op_7826_end_0"), val = tensor([2, 144, 1, 448])]; tensor var_7826_end_mask_0 = const()[name = tensor("op_7826_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7826_cast_fp16 = slice_by_index(begin = var_7826_begin_0, end = var_7826_end_0, end_mask = var_7826_end_mask_0, x = transpose_13)[name = tensor("op_7826_cast_fp16")]; tensor var_7830_begin_0 = const()[name = tensor("op_7830_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_7830_end_0 = const()[name = tensor("op_7830_end_0"), val = tensor([2, 144, 1, 512])]; tensor var_7830_end_mask_0 = const()[name = tensor("op_7830_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7830_cast_fp16 = slice_by_index(begin = var_7830_begin_0, end = var_7830_end_0, end_mask = var_7830_end_mask_0, x = transpose_13)[name = tensor("op_7830_cast_fp16")]; tensor var_7834_begin_0 = const()[name = tensor("op_7834_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7834_end_0 = const()[name = tensor("op_7834_end_0"), val = tensor([2, 144, 1, 576])]; tensor var_7834_end_mask_0 = const()[name = tensor("op_7834_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7834_cast_fp16 = slice_by_index(begin = var_7834_begin_0, end = var_7834_end_0, end_mask = var_7834_end_mask_0, x = transpose_13)[name = tensor("op_7834_cast_fp16")]; tensor var_7838_begin_0 = const()[name = tensor("op_7838_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_7838_end_0 = const()[name = tensor("op_7838_end_0"), val = tensor([2, 144, 1, 640])]; tensor var_7838_end_mask_0 = const()[name = tensor("op_7838_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7838_cast_fp16 = slice_by_index(begin = var_7838_begin_0, end = var_7838_end_0, end_mask = var_7838_end_mask_0, x = transpose_13)[name = tensor("op_7838_cast_fp16")]; tensor var_7842_begin_0 = const()[name = tensor("op_7842_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_7842_end_0 = const()[name = tensor("op_7842_end_0"), val = tensor([2, 144, 1, 704])]; tensor var_7842_end_mask_0 = const()[name = tensor("op_7842_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7842_cast_fp16 = slice_by_index(begin = var_7842_begin_0, end = var_7842_end_0, end_mask = var_7842_end_mask_0, x = transpose_13)[name = tensor("op_7842_cast_fp16")]; tensor var_7846_begin_0 = const()[name = tensor("op_7846_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_7846_end_0 = const()[name = tensor("op_7846_end_0"), val = tensor([2, 144, 1, 768])]; tensor var_7846_end_mask_0 = const()[name = tensor("op_7846_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7846_cast_fp16 = slice_by_index(begin = var_7846_begin_0, end = var_7846_end_0, end_mask = var_7846_end_mask_0, x = transpose_13)[name = tensor("op_7846_cast_fp16")]; tensor var_7850_begin_0 = const()[name = tensor("op_7850_begin_0"), val = tensor([0, 0, 0, 768])]; tensor var_7850_end_0 = const()[name = tensor("op_7850_end_0"), val = tensor([2, 144, 1, 832])]; tensor var_7850_end_mask_0 = const()[name = tensor("op_7850_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7850_cast_fp16 = slice_by_index(begin = var_7850_begin_0, end = var_7850_end_0, end_mask = var_7850_end_mask_0, x = transpose_13)[name = tensor("op_7850_cast_fp16")]; tensor var_7854_begin_0 = const()[name = tensor("op_7854_begin_0"), val = tensor([0, 0, 0, 832])]; tensor var_7854_end_0 = const()[name = tensor("op_7854_end_0"), val = tensor([2, 144, 1, 896])]; tensor var_7854_end_mask_0 = const()[name = tensor("op_7854_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7854_cast_fp16 = slice_by_index(begin = var_7854_begin_0, end = var_7854_end_0, end_mask = var_7854_end_mask_0, x = transpose_13)[name = tensor("op_7854_cast_fp16")]; tensor var_7858_begin_0 = const()[name = tensor("op_7858_begin_0"), val = tensor([0, 0, 0, 896])]; tensor var_7858_end_0 = const()[name = tensor("op_7858_end_0"), val = tensor([2, 144, 1, 960])]; tensor var_7858_end_mask_0 = const()[name = tensor("op_7858_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7858_cast_fp16 = slice_by_index(begin = var_7858_begin_0, end = var_7858_end_0, end_mask = var_7858_end_mask_0, x = transpose_13)[name = tensor("op_7858_cast_fp16")]; tensor var_7862_begin_0 = const()[name = tensor("op_7862_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_7862_end_0 = const()[name = tensor("op_7862_end_0"), val = tensor([2, 144, 1, 1024])]; tensor var_7862_end_mask_0 = const()[name = tensor("op_7862_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7862_cast_fp16 = slice_by_index(begin = var_7862_begin_0, end = var_7862_end_0, end_mask = var_7862_end_mask_0, x = transpose_13)[name = tensor("op_7862_cast_fp16")]; tensor var_7866_begin_0 = const()[name = tensor("op_7866_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7866_end_0 = const()[name = tensor("op_7866_end_0"), val = tensor([2, 144, 1, 1088])]; tensor var_7866_end_mask_0 = const()[name = tensor("op_7866_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7866_cast_fp16 = slice_by_index(begin = var_7866_begin_0, end = var_7866_end_0, end_mask = var_7866_end_mask_0, x = transpose_13)[name = tensor("op_7866_cast_fp16")]; tensor var_7870_begin_0 = const()[name = tensor("op_7870_begin_0"), val = tensor([0, 0, 0, 1088])]; tensor var_7870_end_0 = const()[name = tensor("op_7870_end_0"), val = tensor([2, 144, 1, 1152])]; tensor var_7870_end_mask_0 = const()[name = tensor("op_7870_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7870_cast_fp16 = slice_by_index(begin = var_7870_begin_0, end = var_7870_end_0, end_mask = var_7870_end_mask_0, x = transpose_13)[name = tensor("op_7870_cast_fp16")]; tensor var_7874_begin_0 = const()[name = tensor("op_7874_begin_0"), val = tensor([0, 0, 0, 1152])]; tensor var_7874_end_0 = const()[name = tensor("op_7874_end_0"), val = tensor([2, 144, 1, 1216])]; tensor var_7874_end_mask_0 = const()[name = tensor("op_7874_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7874_cast_fp16 = slice_by_index(begin = var_7874_begin_0, end = var_7874_end_0, end_mask = var_7874_end_mask_0, x = transpose_13)[name = tensor("op_7874_cast_fp16")]; tensor var_7878_begin_0 = const()[name = tensor("op_7878_begin_0"), val = tensor([0, 0, 0, 1216])]; tensor var_7878_end_0 = const()[name = tensor("op_7878_end_0"), val = tensor([2, 144, 1, 1280])]; tensor var_7878_end_mask_0 = const()[name = tensor("op_7878_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7878_cast_fp16 = slice_by_index(begin = var_7878_begin_0, end = var_7878_end_0, end_mask = var_7878_end_mask_0, x = transpose_13)[name = tensor("op_7878_cast_fp16")]; tensor var_7880_begin_0 = const()[name = tensor("op_7880_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7880_end_0 = const()[name = tensor("op_7880_end_0"), val = tensor([2, 64, 1, 144])]; tensor var_7880_end_mask_0 = const()[name = tensor("op_7880_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7880_cast_fp16 = slice_by_index(begin = var_7880_begin_0, end = var_7880_end_0, end_mask = var_7880_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7880_cast_fp16")]; tensor var_7884_begin_0 = const()[name = tensor("op_7884_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_7884_end_0 = const()[name = tensor("op_7884_end_0"), val = tensor([2, 128, 1, 144])]; tensor var_7884_end_mask_0 = const()[name = tensor("op_7884_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7884_cast_fp16 = slice_by_index(begin = var_7884_begin_0, end = var_7884_end_0, end_mask = var_7884_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7884_cast_fp16")]; tensor var_7888_begin_0 = const()[name = tensor("op_7888_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_7888_end_0 = const()[name = tensor("op_7888_end_0"), val = tensor([2, 192, 1, 144])]; tensor var_7888_end_mask_0 = const()[name = tensor("op_7888_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7888_cast_fp16 = slice_by_index(begin = var_7888_begin_0, end = var_7888_end_0, end_mask = var_7888_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7888_cast_fp16")]; tensor var_7892_begin_0 = const()[name = tensor("op_7892_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_7892_end_0 = const()[name = tensor("op_7892_end_0"), val = tensor([2, 256, 1, 144])]; tensor var_7892_end_mask_0 = const()[name = tensor("op_7892_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7892_cast_fp16 = slice_by_index(begin = var_7892_begin_0, end = var_7892_end_0, end_mask = var_7892_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7892_cast_fp16")]; tensor var_7896_begin_0 = const()[name = tensor("op_7896_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_7896_end_0 = const()[name = tensor("op_7896_end_0"), val = tensor([2, 320, 1, 144])]; tensor var_7896_end_mask_0 = const()[name = tensor("op_7896_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7896_cast_fp16 = slice_by_index(begin = var_7896_begin_0, end = var_7896_end_0, end_mask = var_7896_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7896_cast_fp16")]; tensor var_7900_begin_0 = const()[name = tensor("op_7900_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7900_end_0 = const()[name = tensor("op_7900_end_0"), val = tensor([2, 384, 1, 144])]; tensor var_7900_end_mask_0 = const()[name = tensor("op_7900_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7900_cast_fp16 = slice_by_index(begin = var_7900_begin_0, end = var_7900_end_0, end_mask = var_7900_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7900_cast_fp16")]; tensor var_7904_begin_0 = const()[name = tensor("op_7904_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_7904_end_0 = const()[name = tensor("op_7904_end_0"), val = tensor([2, 448, 1, 144])]; tensor var_7904_end_mask_0 = const()[name = tensor("op_7904_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7904_cast_fp16 = slice_by_index(begin = var_7904_begin_0, end = var_7904_end_0, end_mask = var_7904_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7904_cast_fp16")]; tensor var_7908_begin_0 = const()[name = tensor("op_7908_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_7908_end_0 = const()[name = tensor("op_7908_end_0"), val = tensor([2, 512, 1, 144])]; tensor var_7908_end_mask_0 = const()[name = tensor("op_7908_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7908_cast_fp16 = slice_by_index(begin = var_7908_begin_0, end = var_7908_end_0, end_mask = var_7908_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7908_cast_fp16")]; tensor var_7912_begin_0 = const()[name = tensor("op_7912_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_7912_end_0 = const()[name = tensor("op_7912_end_0"), val = tensor([2, 576, 1, 144])]; tensor var_7912_end_mask_0 = const()[name = tensor("op_7912_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7912_cast_fp16 = slice_by_index(begin = var_7912_begin_0, end = var_7912_end_0, end_mask = var_7912_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7912_cast_fp16")]; tensor var_7916_begin_0 = const()[name = tensor("op_7916_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_7916_end_0 = const()[name = tensor("op_7916_end_0"), val = tensor([2, 640, 1, 144])]; tensor var_7916_end_mask_0 = const()[name = tensor("op_7916_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7916_cast_fp16 = slice_by_index(begin = var_7916_begin_0, end = var_7916_end_0, end_mask = var_7916_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7916_cast_fp16")]; tensor var_7920_begin_0 = const()[name = tensor("op_7920_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7920_end_0 = const()[name = tensor("op_7920_end_0"), val = tensor([2, 704, 1, 144])]; tensor var_7920_end_mask_0 = const()[name = tensor("op_7920_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7920_cast_fp16 = slice_by_index(begin = var_7920_begin_0, end = var_7920_end_0, end_mask = var_7920_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7920_cast_fp16")]; tensor var_7924_begin_0 = const()[name = tensor("op_7924_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_7924_end_0 = const()[name = tensor("op_7924_end_0"), val = tensor([2, 768, 1, 144])]; tensor var_7924_end_mask_0 = const()[name = tensor("op_7924_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7924_cast_fp16 = slice_by_index(begin = var_7924_begin_0, end = var_7924_end_0, end_mask = var_7924_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7924_cast_fp16")]; tensor var_7928_begin_0 = const()[name = tensor("op_7928_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_7928_end_0 = const()[name = tensor("op_7928_end_0"), val = tensor([2, 832, 1, 144])]; tensor var_7928_end_mask_0 = const()[name = tensor("op_7928_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7928_cast_fp16 = slice_by_index(begin = var_7928_begin_0, end = var_7928_end_0, end_mask = var_7928_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7928_cast_fp16")]; tensor var_7932_begin_0 = const()[name = tensor("op_7932_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_7932_end_0 = const()[name = tensor("op_7932_end_0"), val = tensor([2, 896, 1, 144])]; tensor var_7932_end_mask_0 = const()[name = tensor("op_7932_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7932_cast_fp16 = slice_by_index(begin = var_7932_begin_0, end = var_7932_end_0, end_mask = var_7932_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7932_cast_fp16")]; tensor var_7936_begin_0 = const()[name = tensor("op_7936_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_7936_end_0 = const()[name = tensor("op_7936_end_0"), val = tensor([2, 960, 1, 144])]; tensor var_7936_end_mask_0 = const()[name = tensor("op_7936_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7936_cast_fp16 = slice_by_index(begin = var_7936_begin_0, end = var_7936_end_0, end_mask = var_7936_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7936_cast_fp16")]; tensor var_7940_begin_0 = const()[name = tensor("op_7940_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_7940_end_0 = const()[name = tensor("op_7940_end_0"), val = tensor([2, 1024, 1, 144])]; tensor var_7940_end_mask_0 = const()[name = tensor("op_7940_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7940_cast_fp16 = slice_by_index(begin = var_7940_begin_0, end = var_7940_end_0, end_mask = var_7940_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7940_cast_fp16")]; tensor var_7944_begin_0 = const()[name = tensor("op_7944_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_7944_end_0 = const()[name = tensor("op_7944_end_0"), val = tensor([2, 1088, 1, 144])]; tensor var_7944_end_mask_0 = const()[name = tensor("op_7944_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7944_cast_fp16 = slice_by_index(begin = var_7944_begin_0, end = var_7944_end_0, end_mask = var_7944_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7944_cast_fp16")]; tensor var_7948_begin_0 = const()[name = tensor("op_7948_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_7948_end_0 = const()[name = tensor("op_7948_end_0"), val = tensor([2, 1152, 1, 144])]; tensor var_7948_end_mask_0 = const()[name = tensor("op_7948_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7948_cast_fp16 = slice_by_index(begin = var_7948_begin_0, end = var_7948_end_0, end_mask = var_7948_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7948_cast_fp16")]; tensor var_7952_begin_0 = const()[name = tensor("op_7952_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_7952_end_0 = const()[name = tensor("op_7952_end_0"), val = tensor([2, 1216, 1, 144])]; tensor var_7952_end_mask_0 = const()[name = tensor("op_7952_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7952_cast_fp16 = slice_by_index(begin = var_7952_begin_0, end = var_7952_end_0, end_mask = var_7952_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7952_cast_fp16")]; tensor var_7956_begin_0 = const()[name = tensor("op_7956_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_7956_end_0 = const()[name = tensor("op_7956_end_0"), val = tensor([2, 1280, 1, 144])]; tensor var_7956_end_mask_0 = const()[name = tensor("op_7956_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7956_cast_fp16 = slice_by_index(begin = var_7956_begin_0, end = var_7956_end_0, end_mask = var_7956_end_mask_0, x = v_37_cast_fp16)[name = tensor("op_7956_cast_fp16")]; tensor var_7960_equation_0 = const()[name = tensor("op_7960_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7960_cast_fp16 = einsum(equation = var_7960_equation_0, values = (var_7802_cast_fp16, var_7719_cast_fp16))[name = tensor("op_7960_cast_fp16")]; tensor var_7961_to_fp16 = const()[name = tensor("op_7961_to_fp16"), val = tensor(0x1p-3)]; tensor aw_521_cast_fp16 = mul(x = var_7960_cast_fp16, y = var_7961_to_fp16)[name = tensor("aw_521_cast_fp16")]; tensor var_7964_equation_0 = const()[name = tensor("op_7964_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7964_cast_fp16 = einsum(equation = var_7964_equation_0, values = (var_7806_cast_fp16, var_7723_cast_fp16))[name = tensor("op_7964_cast_fp16")]; tensor var_7965_to_fp16 = const()[name = tensor("op_7965_to_fp16"), val = tensor(0x1p-3)]; tensor aw_523_cast_fp16 = mul(x = var_7964_cast_fp16, y = var_7965_to_fp16)[name = tensor("aw_523_cast_fp16")]; tensor var_7968_equation_0 = const()[name = tensor("op_7968_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7968_cast_fp16 = einsum(equation = var_7968_equation_0, values = (var_7810_cast_fp16, var_7727_cast_fp16))[name = tensor("op_7968_cast_fp16")]; tensor var_7969_to_fp16 = const()[name = tensor("op_7969_to_fp16"), val = tensor(0x1p-3)]; tensor aw_525_cast_fp16 = mul(x = var_7968_cast_fp16, y = var_7969_to_fp16)[name = tensor("aw_525_cast_fp16")]; tensor var_7972_equation_0 = const()[name = tensor("op_7972_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7972_cast_fp16 = einsum(equation = var_7972_equation_0, values = (var_7814_cast_fp16, var_7731_cast_fp16))[name = tensor("op_7972_cast_fp16")]; tensor var_7973_to_fp16 = const()[name = tensor("op_7973_to_fp16"), val = tensor(0x1p-3)]; tensor aw_527_cast_fp16 = mul(x = var_7972_cast_fp16, y = var_7973_to_fp16)[name = tensor("aw_527_cast_fp16")]; tensor var_7976_equation_0 = const()[name = tensor("op_7976_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7976_cast_fp16 = einsum(equation = var_7976_equation_0, values = (var_7818_cast_fp16, var_7735_cast_fp16))[name = tensor("op_7976_cast_fp16")]; tensor var_7977_to_fp16 = const()[name = tensor("op_7977_to_fp16"), val = tensor(0x1p-3)]; tensor aw_529_cast_fp16 = mul(x = var_7976_cast_fp16, y = var_7977_to_fp16)[name = tensor("aw_529_cast_fp16")]; tensor var_7980_equation_0 = const()[name = tensor("op_7980_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7980_cast_fp16 = einsum(equation = var_7980_equation_0, values = (var_7822_cast_fp16, var_7739_cast_fp16))[name = tensor("op_7980_cast_fp16")]; tensor var_7981_to_fp16 = const()[name = tensor("op_7981_to_fp16"), val = tensor(0x1p-3)]; tensor aw_531_cast_fp16 = mul(x = var_7980_cast_fp16, y = var_7981_to_fp16)[name = tensor("aw_531_cast_fp16")]; tensor var_7984_equation_0 = const()[name = tensor("op_7984_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7984_cast_fp16 = einsum(equation = var_7984_equation_0, values = (var_7826_cast_fp16, var_7743_cast_fp16))[name = tensor("op_7984_cast_fp16")]; tensor var_7985_to_fp16 = const()[name = tensor("op_7985_to_fp16"), val = tensor(0x1p-3)]; tensor aw_533_cast_fp16 = mul(x = var_7984_cast_fp16, y = var_7985_to_fp16)[name = tensor("aw_533_cast_fp16")]; tensor var_7988_equation_0 = const()[name = tensor("op_7988_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7988_cast_fp16 = einsum(equation = var_7988_equation_0, values = (var_7830_cast_fp16, var_7747_cast_fp16))[name = tensor("op_7988_cast_fp16")]; tensor var_7989_to_fp16 = const()[name = tensor("op_7989_to_fp16"), val = tensor(0x1p-3)]; tensor aw_535_cast_fp16 = mul(x = var_7988_cast_fp16, y = var_7989_to_fp16)[name = tensor("aw_535_cast_fp16")]; tensor var_7992_equation_0 = const()[name = tensor("op_7992_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7992_cast_fp16 = einsum(equation = var_7992_equation_0, values = (var_7834_cast_fp16, var_7751_cast_fp16))[name = tensor("op_7992_cast_fp16")]; tensor var_7993_to_fp16 = const()[name = tensor("op_7993_to_fp16"), val = tensor(0x1p-3)]; tensor aw_537_cast_fp16 = mul(x = var_7992_cast_fp16, y = var_7993_to_fp16)[name = tensor("aw_537_cast_fp16")]; tensor var_7996_equation_0 = const()[name = tensor("op_7996_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7996_cast_fp16 = einsum(equation = var_7996_equation_0, values = (var_7838_cast_fp16, var_7755_cast_fp16))[name = tensor("op_7996_cast_fp16")]; tensor var_7997_to_fp16 = const()[name = tensor("op_7997_to_fp16"), val = tensor(0x1p-3)]; tensor aw_539_cast_fp16 = mul(x = var_7996_cast_fp16, y = var_7997_to_fp16)[name = tensor("aw_539_cast_fp16")]; tensor var_8000_equation_0 = const()[name = tensor("op_8000_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8000_cast_fp16 = einsum(equation = var_8000_equation_0, values = (var_7842_cast_fp16, var_7759_cast_fp16))[name = tensor("op_8000_cast_fp16")]; tensor var_8001_to_fp16 = const()[name = tensor("op_8001_to_fp16"), val = tensor(0x1p-3)]; tensor aw_541_cast_fp16 = mul(x = var_8000_cast_fp16, y = var_8001_to_fp16)[name = tensor("aw_541_cast_fp16")]; tensor var_8004_equation_0 = const()[name = tensor("op_8004_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8004_cast_fp16 = einsum(equation = var_8004_equation_0, values = (var_7846_cast_fp16, var_7763_cast_fp16))[name = tensor("op_8004_cast_fp16")]; tensor var_8005_to_fp16 = const()[name = tensor("op_8005_to_fp16"), val = tensor(0x1p-3)]; tensor aw_543_cast_fp16 = mul(x = var_8004_cast_fp16, y = var_8005_to_fp16)[name = tensor("aw_543_cast_fp16")]; tensor var_8008_equation_0 = const()[name = tensor("op_8008_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8008_cast_fp16 = einsum(equation = var_8008_equation_0, values = (var_7850_cast_fp16, var_7767_cast_fp16))[name = tensor("op_8008_cast_fp16")]; tensor var_8009_to_fp16 = const()[name = tensor("op_8009_to_fp16"), val = tensor(0x1p-3)]; tensor aw_545_cast_fp16 = mul(x = var_8008_cast_fp16, y = var_8009_to_fp16)[name = tensor("aw_545_cast_fp16")]; tensor var_8012_equation_0 = const()[name = tensor("op_8012_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8012_cast_fp16 = einsum(equation = var_8012_equation_0, values = (var_7854_cast_fp16, var_7771_cast_fp16))[name = tensor("op_8012_cast_fp16")]; tensor var_8013_to_fp16 = const()[name = tensor("op_8013_to_fp16"), val = tensor(0x1p-3)]; tensor aw_547_cast_fp16 = mul(x = var_8012_cast_fp16, y = var_8013_to_fp16)[name = tensor("aw_547_cast_fp16")]; tensor var_8016_equation_0 = const()[name = tensor("op_8016_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8016_cast_fp16 = einsum(equation = var_8016_equation_0, values = (var_7858_cast_fp16, var_7775_cast_fp16))[name = tensor("op_8016_cast_fp16")]; tensor var_8017_to_fp16 = const()[name = tensor("op_8017_to_fp16"), val = tensor(0x1p-3)]; tensor aw_549_cast_fp16 = mul(x = var_8016_cast_fp16, y = var_8017_to_fp16)[name = tensor("aw_549_cast_fp16")]; tensor var_8020_equation_0 = const()[name = tensor("op_8020_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8020_cast_fp16 = einsum(equation = var_8020_equation_0, values = (var_7862_cast_fp16, var_7779_cast_fp16))[name = tensor("op_8020_cast_fp16")]; tensor var_8021_to_fp16 = const()[name = tensor("op_8021_to_fp16"), val = tensor(0x1p-3)]; tensor aw_551_cast_fp16 = mul(x = var_8020_cast_fp16, y = var_8021_to_fp16)[name = tensor("aw_551_cast_fp16")]; tensor var_8024_equation_0 = const()[name = tensor("op_8024_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8024_cast_fp16 = einsum(equation = var_8024_equation_0, values = (var_7866_cast_fp16, var_7783_cast_fp16))[name = tensor("op_8024_cast_fp16")]; tensor var_8025_to_fp16 = const()[name = tensor("op_8025_to_fp16"), val = tensor(0x1p-3)]; tensor aw_553_cast_fp16 = mul(x = var_8024_cast_fp16, y = var_8025_to_fp16)[name = tensor("aw_553_cast_fp16")]; tensor var_8028_equation_0 = const()[name = tensor("op_8028_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8028_cast_fp16 = einsum(equation = var_8028_equation_0, values = (var_7870_cast_fp16, var_7787_cast_fp16))[name = tensor("op_8028_cast_fp16")]; tensor var_8029_to_fp16 = const()[name = tensor("op_8029_to_fp16"), val = tensor(0x1p-3)]; tensor aw_555_cast_fp16 = mul(x = var_8028_cast_fp16, y = var_8029_to_fp16)[name = tensor("aw_555_cast_fp16")]; tensor var_8032_equation_0 = const()[name = tensor("op_8032_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8032_cast_fp16 = einsum(equation = var_8032_equation_0, values = (var_7874_cast_fp16, var_7791_cast_fp16))[name = tensor("op_8032_cast_fp16")]; tensor var_8033_to_fp16 = const()[name = tensor("op_8033_to_fp16"), val = tensor(0x1p-3)]; tensor aw_557_cast_fp16 = mul(x = var_8032_cast_fp16, y = var_8033_to_fp16)[name = tensor("aw_557_cast_fp16")]; tensor var_8036_equation_0 = const()[name = tensor("op_8036_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8036_cast_fp16 = einsum(equation = var_8036_equation_0, values = (var_7878_cast_fp16, var_7795_cast_fp16))[name = tensor("op_8036_cast_fp16")]; tensor var_8037_to_fp16 = const()[name = tensor("op_8037_to_fp16"), val = tensor(0x1p-3)]; tensor aw_559_cast_fp16 = mul(x = var_8036_cast_fp16, y = var_8037_to_fp16)[name = tensor("aw_559_cast_fp16")]; tensor var_8039_cast_fp16 = softmax(axis = var_5599, x = aw_521_cast_fp16)[name = tensor("op_8039_cast_fp16")]; tensor var_8040_cast_fp16 = softmax(axis = var_5599, x = aw_523_cast_fp16)[name = tensor("op_8040_cast_fp16")]; tensor var_8041_cast_fp16 = softmax(axis = var_5599, x = aw_525_cast_fp16)[name = tensor("op_8041_cast_fp16")]; tensor var_8042_cast_fp16 = softmax(axis = var_5599, x = aw_527_cast_fp16)[name = tensor("op_8042_cast_fp16")]; tensor var_8043_cast_fp16 = softmax(axis = var_5599, x = aw_529_cast_fp16)[name = tensor("op_8043_cast_fp16")]; tensor var_8044_cast_fp16 = softmax(axis = var_5599, x = aw_531_cast_fp16)[name = tensor("op_8044_cast_fp16")]; tensor var_8045_cast_fp16 = softmax(axis = var_5599, x = aw_533_cast_fp16)[name = tensor("op_8045_cast_fp16")]; tensor var_8046_cast_fp16 = softmax(axis = var_5599, x = aw_535_cast_fp16)[name = tensor("op_8046_cast_fp16")]; tensor var_8047_cast_fp16 = softmax(axis = var_5599, x = aw_537_cast_fp16)[name = tensor("op_8047_cast_fp16")]; tensor var_8048_cast_fp16 = softmax(axis = var_5599, x = aw_539_cast_fp16)[name = tensor("op_8048_cast_fp16")]; tensor var_8049_cast_fp16 = softmax(axis = var_5599, x = aw_541_cast_fp16)[name = tensor("op_8049_cast_fp16")]; tensor var_8050_cast_fp16 = softmax(axis = var_5599, x = aw_543_cast_fp16)[name = tensor("op_8050_cast_fp16")]; tensor var_8051_cast_fp16 = softmax(axis = var_5599, x = aw_545_cast_fp16)[name = tensor("op_8051_cast_fp16")]; tensor var_8052_cast_fp16 = softmax(axis = var_5599, x = aw_547_cast_fp16)[name = tensor("op_8052_cast_fp16")]; tensor var_8053_cast_fp16 = softmax(axis = var_5599, x = aw_549_cast_fp16)[name = tensor("op_8053_cast_fp16")]; tensor var_8054_cast_fp16 = softmax(axis = var_5599, x = aw_551_cast_fp16)[name = tensor("op_8054_cast_fp16")]; tensor var_8055_cast_fp16 = softmax(axis = var_5599, x = aw_553_cast_fp16)[name = tensor("op_8055_cast_fp16")]; tensor var_8056_cast_fp16 = softmax(axis = var_5599, x = aw_555_cast_fp16)[name = tensor("op_8056_cast_fp16")]; tensor var_8057_cast_fp16 = softmax(axis = var_5599, x = aw_557_cast_fp16)[name = tensor("op_8057_cast_fp16")]; tensor var_8058_cast_fp16 = softmax(axis = var_5599, x = aw_559_cast_fp16)[name = tensor("op_8058_cast_fp16")]; tensor var_8060_equation_0 = const()[name = tensor("op_8060_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8060_cast_fp16 = einsum(equation = var_8060_equation_0, values = (var_7880_cast_fp16, var_8039_cast_fp16))[name = tensor("op_8060_cast_fp16")]; tensor var_8062_equation_0 = const()[name = tensor("op_8062_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8062_cast_fp16 = einsum(equation = var_8062_equation_0, values = (var_7884_cast_fp16, var_8040_cast_fp16))[name = tensor("op_8062_cast_fp16")]; tensor var_8064_equation_0 = const()[name = tensor("op_8064_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8064_cast_fp16 = einsum(equation = var_8064_equation_0, values = (var_7888_cast_fp16, var_8041_cast_fp16))[name = tensor("op_8064_cast_fp16")]; tensor var_8066_equation_0 = const()[name = tensor("op_8066_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8066_cast_fp16 = einsum(equation = var_8066_equation_0, values = (var_7892_cast_fp16, var_8042_cast_fp16))[name = tensor("op_8066_cast_fp16")]; tensor var_8068_equation_0 = const()[name = tensor("op_8068_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8068_cast_fp16 = einsum(equation = var_8068_equation_0, values = (var_7896_cast_fp16, var_8043_cast_fp16))[name = tensor("op_8068_cast_fp16")]; tensor var_8070_equation_0 = const()[name = tensor("op_8070_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8070_cast_fp16 = einsum(equation = var_8070_equation_0, values = (var_7900_cast_fp16, var_8044_cast_fp16))[name = tensor("op_8070_cast_fp16")]; tensor var_8072_equation_0 = const()[name = tensor("op_8072_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8072_cast_fp16 = einsum(equation = var_8072_equation_0, values = (var_7904_cast_fp16, var_8045_cast_fp16))[name = tensor("op_8072_cast_fp16")]; tensor var_8074_equation_0 = const()[name = tensor("op_8074_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8074_cast_fp16 = einsum(equation = var_8074_equation_0, values = (var_7908_cast_fp16, var_8046_cast_fp16))[name = tensor("op_8074_cast_fp16")]; tensor var_8076_equation_0 = const()[name = tensor("op_8076_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8076_cast_fp16 = einsum(equation = var_8076_equation_0, values = (var_7912_cast_fp16, var_8047_cast_fp16))[name = tensor("op_8076_cast_fp16")]; tensor var_8078_equation_0 = const()[name = tensor("op_8078_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8078_cast_fp16 = einsum(equation = var_8078_equation_0, values = (var_7916_cast_fp16, var_8048_cast_fp16))[name = tensor("op_8078_cast_fp16")]; tensor var_8080_equation_0 = const()[name = tensor("op_8080_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8080_cast_fp16 = einsum(equation = var_8080_equation_0, values = (var_7920_cast_fp16, var_8049_cast_fp16))[name = tensor("op_8080_cast_fp16")]; tensor var_8082_equation_0 = const()[name = tensor("op_8082_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8082_cast_fp16 = einsum(equation = var_8082_equation_0, values = (var_7924_cast_fp16, var_8050_cast_fp16))[name = tensor("op_8082_cast_fp16")]; tensor var_8084_equation_0 = const()[name = tensor("op_8084_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8084_cast_fp16 = einsum(equation = var_8084_equation_0, values = (var_7928_cast_fp16, var_8051_cast_fp16))[name = tensor("op_8084_cast_fp16")]; tensor var_8086_equation_0 = const()[name = tensor("op_8086_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8086_cast_fp16 = einsum(equation = var_8086_equation_0, values = (var_7932_cast_fp16, var_8052_cast_fp16))[name = tensor("op_8086_cast_fp16")]; tensor var_8088_equation_0 = const()[name = tensor("op_8088_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8088_cast_fp16 = einsum(equation = var_8088_equation_0, values = (var_7936_cast_fp16, var_8053_cast_fp16))[name = tensor("op_8088_cast_fp16")]; tensor var_8090_equation_0 = const()[name = tensor("op_8090_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8090_cast_fp16 = einsum(equation = var_8090_equation_0, values = (var_7940_cast_fp16, var_8054_cast_fp16))[name = tensor("op_8090_cast_fp16")]; tensor var_8092_equation_0 = const()[name = tensor("op_8092_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8092_cast_fp16 = einsum(equation = var_8092_equation_0, values = (var_7944_cast_fp16, var_8055_cast_fp16))[name = tensor("op_8092_cast_fp16")]; tensor var_8094_equation_0 = const()[name = tensor("op_8094_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8094_cast_fp16 = einsum(equation = var_8094_equation_0, values = (var_7948_cast_fp16, var_8056_cast_fp16))[name = tensor("op_8094_cast_fp16")]; tensor var_8096_equation_0 = const()[name = tensor("op_8096_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8096_cast_fp16 = einsum(equation = var_8096_equation_0, values = (var_7952_cast_fp16, var_8057_cast_fp16))[name = tensor("op_8096_cast_fp16")]; tensor var_8098_equation_0 = const()[name = tensor("op_8098_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8098_cast_fp16 = einsum(equation = var_8098_equation_0, values = (var_7956_cast_fp16, var_8058_cast_fp16))[name = tensor("op_8098_cast_fp16")]; tensor input_353_interleave_0 = const()[name = tensor("input_353_interleave_0"), val = tensor(false)]; tensor input_353_cast_fp16 = concat(axis = var_5599, interleave = input_353_interleave_0, values = (var_8060_cast_fp16, var_8062_cast_fp16, var_8064_cast_fp16, var_8066_cast_fp16, var_8068_cast_fp16, var_8070_cast_fp16, var_8072_cast_fp16, var_8074_cast_fp16, var_8076_cast_fp16, var_8078_cast_fp16, var_8080_cast_fp16, var_8082_cast_fp16, var_8084_cast_fp16, var_8086_cast_fp16, var_8088_cast_fp16, var_8090_cast_fp16, var_8092_cast_fp16, var_8094_cast_fp16, var_8096_cast_fp16, var_8098_cast_fp16))[name = tensor("input_353_cast_fp16")]; tensor var_8104 = const()[name = tensor("op_8104"), val = tensor([1, 1])]; tensor var_8106 = const()[name = tensor("op_8106"), val = tensor([1, 1])]; tensor var_8108_pad_type_0 = const()[name = tensor("op_8108_pad_type_0"), val = tensor("custom")]; tensor var_8108_pad_0 = const()[name = tensor("op_8108_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(567801856)))]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571078720)))]; tensor var_8108_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_8106, groups = var_5599, pad = var_8108_pad_0, pad_type = var_8108_pad_type_0, strides = var_8104, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_353_cast_fp16)[name = tensor("op_8108_cast_fp16")]; tensor inputs_57_cast_fp16 = add(x = var_8108_cast_fp16, y = inputs_55_cast_fp16)[name = tensor("inputs_57_cast_fp16")]; tensor var_8112 = const()[name = tensor("op_8112"), val = tensor([1])]; tensor channels_mean_57_cast_fp16 = reduce_mean(axes = var_8112, keep_dims = var_5594, x = inputs_57_cast_fp16)[name = tensor("channels_mean_57_cast_fp16")]; tensor zero_mean_57_cast_fp16 = sub(x = inputs_57_cast_fp16, y = channels_mean_57_cast_fp16)[name = tensor("zero_mean_57_cast_fp16")]; tensor zero_mean_sq_57_cast_fp16 = mul(x = zero_mean_57_cast_fp16, y = zero_mean_57_cast_fp16)[name = tensor("zero_mean_sq_57_cast_fp16")]; tensor var_8116 = const()[name = tensor("op_8116"), val = tensor([1])]; tensor var_8117_cast_fp16 = reduce_mean(axes = var_8116, keep_dims = var_5594, x = zero_mean_sq_57_cast_fp16)[name = tensor("op_8117_cast_fp16")]; tensor var_8118_to_fp16 = const()[name = tensor("op_8118_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_8119_cast_fp16 = add(x = var_8117_cast_fp16, y = var_8118_to_fp16)[name = tensor("op_8119_cast_fp16")]; tensor denom_57_epsilon_0_to_fp16 = const()[name = tensor("denom_57_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_57_cast_fp16 = rsqrt(epsilon = denom_57_epsilon_0_to_fp16, x = var_8119_cast_fp16)[name = tensor("denom_57_cast_fp16")]; tensor out_57_cast_fp16 = mul(x = zero_mean_57_cast_fp16, y = denom_57_cast_fp16)[name = tensor("out_57_cast_fp16")]; tensor var_8123_to_fp16 = const()[name = tensor("op_8123_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571081344)))]; tensor var_8124_cast_fp16 = add(x = out_57_cast_fp16, y = var_8123_to_fp16)[name = tensor("op_8124_cast_fp16")]; tensor var_8126_to_fp16 = const()[name = tensor("op_8126_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571083968)))]; tensor hidden_states_209_cast_fp16 = mul(x = var_8124_cast_fp16, y = var_8126_to_fp16)[name = tensor("hidden_states_209_cast_fp16")]; tensor var_8133 = const()[name = tensor("op_8133"), val = tensor([1, 1])]; tensor var_8135 = const()[name = tensor("op_8135"), val = tensor([1, 1])]; tensor q_39_pad_type_0 = const()[name = tensor("q_39_pad_type_0"), val = tensor("custom")]; tensor q_39_pad_0 = const()[name = tensor("q_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571086592)))]; tensor q_39_cast_fp16 = conv(dilations = var_8135, groups = var_5599, pad = q_39_pad_0, pad_type = q_39_pad_type_0, strides = var_8133, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_209_cast_fp16)[name = tensor("q_39_cast_fp16")]; tensor var_8139 = const()[name = tensor("op_8139"), val = tensor([1, 1])]; tensor var_8141 = const()[name = tensor("op_8141"), val = tensor([1, 1])]; tensor k_77_pad_type_0 = const()[name = tensor("k_77_pad_type_0"), val = tensor("custom")]; tensor k_77_pad_0 = const()[name = tensor("k_77_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(574363456)))]; tensor k_77_cast_fp16 = conv(dilations = var_8141, groups = var_5599, pad = k_77_pad_0, pad_type = k_77_pad_type_0, strides = var_8139, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_77_cast_fp16")]; tensor var_8145 = const()[name = tensor("op_8145"), val = tensor([1, 1])]; tensor var_8147 = const()[name = tensor("op_8147"), val = tensor([1, 1])]; tensor v_39_pad_type_0 = const()[name = tensor("v_39_pad_type_0"), val = tensor("custom")]; tensor v_39_pad_0 = const()[name = tensor("v_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(576984960)))]; tensor v_39_cast_fp16 = conv(dilations = var_8147, groups = var_5599, pad = v_39_pad_0, pad_type = v_39_pad_type_0, strides = var_8145, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_39_cast_fp16")]; tensor var_8151_begin_0 = const()[name = tensor("op_8151_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8151_end_0 = const()[name = tensor("op_8151_end_0"), val = tensor([2, 64, 1, 144])]; tensor var_8151_end_mask_0 = const()[name = tensor("op_8151_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8151_cast_fp16 = slice_by_index(begin = var_8151_begin_0, end = var_8151_end_0, end_mask = var_8151_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8151_cast_fp16")]; tensor var_8155_begin_0 = const()[name = tensor("op_8155_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_8155_end_0 = const()[name = tensor("op_8155_end_0"), val = tensor([2, 128, 1, 144])]; tensor var_8155_end_mask_0 = const()[name = tensor("op_8155_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8155_cast_fp16 = slice_by_index(begin = var_8155_begin_0, end = var_8155_end_0, end_mask = var_8155_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8155_cast_fp16")]; tensor var_8159_begin_0 = const()[name = tensor("op_8159_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_8159_end_0 = const()[name = tensor("op_8159_end_0"), val = tensor([2, 192, 1, 144])]; tensor var_8159_end_mask_0 = const()[name = tensor("op_8159_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8159_cast_fp16 = slice_by_index(begin = var_8159_begin_0, end = var_8159_end_0, end_mask = var_8159_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8159_cast_fp16")]; tensor var_8163_begin_0 = const()[name = tensor("op_8163_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_8163_end_0 = const()[name = tensor("op_8163_end_0"), val = tensor([2, 256, 1, 144])]; tensor var_8163_end_mask_0 = const()[name = tensor("op_8163_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8163_cast_fp16 = slice_by_index(begin = var_8163_begin_0, end = var_8163_end_0, end_mask = var_8163_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8163_cast_fp16")]; tensor var_8167_begin_0 = const()[name = tensor("op_8167_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_8167_end_0 = const()[name = tensor("op_8167_end_0"), val = tensor([2, 320, 1, 144])]; tensor var_8167_end_mask_0 = const()[name = tensor("op_8167_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8167_cast_fp16 = slice_by_index(begin = var_8167_begin_0, end = var_8167_end_0, end_mask = var_8167_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8167_cast_fp16")]; tensor var_8171_begin_0 = const()[name = tensor("op_8171_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8171_end_0 = const()[name = tensor("op_8171_end_0"), val = tensor([2, 384, 1, 144])]; tensor var_8171_end_mask_0 = const()[name = tensor("op_8171_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8171_cast_fp16 = slice_by_index(begin = var_8171_begin_0, end = var_8171_end_0, end_mask = var_8171_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8171_cast_fp16")]; tensor var_8175_begin_0 = const()[name = tensor("op_8175_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_8175_end_0 = const()[name = tensor("op_8175_end_0"), val = tensor([2, 448, 1, 144])]; tensor var_8175_end_mask_0 = const()[name = tensor("op_8175_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8175_cast_fp16 = slice_by_index(begin = var_8175_begin_0, end = var_8175_end_0, end_mask = var_8175_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8175_cast_fp16")]; tensor var_8179_begin_0 = const()[name = tensor("op_8179_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_8179_end_0 = const()[name = tensor("op_8179_end_0"), val = tensor([2, 512, 1, 144])]; tensor var_8179_end_mask_0 = const()[name = tensor("op_8179_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8179_cast_fp16 = slice_by_index(begin = var_8179_begin_0, end = var_8179_end_0, end_mask = var_8179_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8179_cast_fp16")]; tensor var_8183_begin_0 = const()[name = tensor("op_8183_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_8183_end_0 = const()[name = tensor("op_8183_end_0"), val = tensor([2, 576, 1, 144])]; tensor var_8183_end_mask_0 = const()[name = tensor("op_8183_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8183_cast_fp16 = slice_by_index(begin = var_8183_begin_0, end = var_8183_end_0, end_mask = var_8183_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8183_cast_fp16")]; tensor var_8187_begin_0 = const()[name = tensor("op_8187_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_8187_end_0 = const()[name = tensor("op_8187_end_0"), val = tensor([2, 640, 1, 144])]; tensor var_8187_end_mask_0 = const()[name = tensor("op_8187_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8187_cast_fp16 = slice_by_index(begin = var_8187_begin_0, end = var_8187_end_0, end_mask = var_8187_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8187_cast_fp16")]; tensor var_8191_begin_0 = const()[name = tensor("op_8191_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_8191_end_0 = const()[name = tensor("op_8191_end_0"), val = tensor([2, 704, 1, 144])]; tensor var_8191_end_mask_0 = const()[name = tensor("op_8191_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8191_cast_fp16 = slice_by_index(begin = var_8191_begin_0, end = var_8191_end_0, end_mask = var_8191_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8191_cast_fp16")]; tensor var_8195_begin_0 = const()[name = tensor("op_8195_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_8195_end_0 = const()[name = tensor("op_8195_end_0"), val = tensor([2, 768, 1, 144])]; tensor var_8195_end_mask_0 = const()[name = tensor("op_8195_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8195_cast_fp16 = slice_by_index(begin = var_8195_begin_0, end = var_8195_end_0, end_mask = var_8195_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8195_cast_fp16")]; tensor var_8199_begin_0 = const()[name = tensor("op_8199_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_8199_end_0 = const()[name = tensor("op_8199_end_0"), val = tensor([2, 832, 1, 144])]; tensor var_8199_end_mask_0 = const()[name = tensor("op_8199_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8199_cast_fp16 = slice_by_index(begin = var_8199_begin_0, end = var_8199_end_0, end_mask = var_8199_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8199_cast_fp16")]; tensor var_8203_begin_0 = const()[name = tensor("op_8203_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_8203_end_0 = const()[name = tensor("op_8203_end_0"), val = tensor([2, 896, 1, 144])]; tensor var_8203_end_mask_0 = const()[name = tensor("op_8203_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8203_cast_fp16 = slice_by_index(begin = var_8203_begin_0, end = var_8203_end_0, end_mask = var_8203_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8203_cast_fp16")]; tensor var_8207_begin_0 = const()[name = tensor("op_8207_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_8207_end_0 = const()[name = tensor("op_8207_end_0"), val = tensor([2, 960, 1, 144])]; tensor var_8207_end_mask_0 = const()[name = tensor("op_8207_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8207_cast_fp16 = slice_by_index(begin = var_8207_begin_0, end = var_8207_end_0, end_mask = var_8207_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8207_cast_fp16")]; tensor var_8211_begin_0 = const()[name = tensor("op_8211_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_8211_end_0 = const()[name = tensor("op_8211_end_0"), val = tensor([2, 1024, 1, 144])]; tensor var_8211_end_mask_0 = const()[name = tensor("op_8211_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8211_cast_fp16 = slice_by_index(begin = var_8211_begin_0, end = var_8211_end_0, end_mask = var_8211_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8211_cast_fp16")]; tensor var_8215_begin_0 = const()[name = tensor("op_8215_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_8215_end_0 = const()[name = tensor("op_8215_end_0"), val = tensor([2, 1088, 1, 144])]; tensor var_8215_end_mask_0 = const()[name = tensor("op_8215_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8215_cast_fp16 = slice_by_index(begin = var_8215_begin_0, end = var_8215_end_0, end_mask = var_8215_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8215_cast_fp16")]; tensor var_8219_begin_0 = const()[name = tensor("op_8219_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_8219_end_0 = const()[name = tensor("op_8219_end_0"), val = tensor([2, 1152, 1, 144])]; tensor var_8219_end_mask_0 = const()[name = tensor("op_8219_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8219_cast_fp16 = slice_by_index(begin = var_8219_begin_0, end = var_8219_end_0, end_mask = var_8219_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8219_cast_fp16")]; tensor var_8223_begin_0 = const()[name = tensor("op_8223_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_8223_end_0 = const()[name = tensor("op_8223_end_0"), val = tensor([2, 1216, 1, 144])]; tensor var_8223_end_mask_0 = const()[name = tensor("op_8223_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8223_cast_fp16 = slice_by_index(begin = var_8223_begin_0, end = var_8223_end_0, end_mask = var_8223_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8223_cast_fp16")]; tensor var_8227_begin_0 = const()[name = tensor("op_8227_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_8227_end_0 = const()[name = tensor("op_8227_end_0"), val = tensor([2, 1280, 1, 144])]; tensor var_8227_end_mask_0 = const()[name = tensor("op_8227_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8227_cast_fp16 = slice_by_index(begin = var_8227_begin_0, end = var_8227_end_0, end_mask = var_8227_end_mask_0, x = q_39_cast_fp16)[name = tensor("op_8227_cast_fp16")]; tensor k_79_perm_0 = const()[name = tensor("k_79_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_8234_begin_0 = const()[name = tensor("op_8234_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8234_end_0 = const()[name = tensor("op_8234_end_0"), val = tensor([2, 77, 1, 64])]; tensor var_8234_end_mask_0 = const()[name = tensor("op_8234_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_12 = transpose(perm = k_79_perm_0, x = k_77_cast_fp16)[name = tensor("transpose_12")]; tensor var_8234_cast_fp16 = slice_by_index(begin = var_8234_begin_0, end = var_8234_end_0, end_mask = var_8234_end_mask_0, x = transpose_12)[name = tensor("op_8234_cast_fp16")]; tensor var_8238_begin_0 = const()[name = tensor("op_8238_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_8238_end_0 = const()[name = tensor("op_8238_end_0"), val = tensor([2, 77, 1, 128])]; tensor var_8238_end_mask_0 = const()[name = tensor("op_8238_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8238_cast_fp16 = slice_by_index(begin = var_8238_begin_0, end = var_8238_end_0, end_mask = var_8238_end_mask_0, x = transpose_12)[name = tensor("op_8238_cast_fp16")]; tensor var_8242_begin_0 = const()[name = tensor("op_8242_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_8242_end_0 = const()[name = tensor("op_8242_end_0"), val = tensor([2, 77, 1, 192])]; tensor var_8242_end_mask_0 = const()[name = tensor("op_8242_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8242_cast_fp16 = slice_by_index(begin = var_8242_begin_0, end = var_8242_end_0, end_mask = var_8242_end_mask_0, x = transpose_12)[name = tensor("op_8242_cast_fp16")]; tensor var_8246_begin_0 = const()[name = tensor("op_8246_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_8246_end_0 = const()[name = tensor("op_8246_end_0"), val = tensor([2, 77, 1, 256])]; tensor var_8246_end_mask_0 = const()[name = tensor("op_8246_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8246_cast_fp16 = slice_by_index(begin = var_8246_begin_0, end = var_8246_end_0, end_mask = var_8246_end_mask_0, x = transpose_12)[name = tensor("op_8246_cast_fp16")]; tensor var_8250_begin_0 = const()[name = tensor("op_8250_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_8250_end_0 = const()[name = tensor("op_8250_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_8250_end_mask_0 = const()[name = tensor("op_8250_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8250_cast_fp16 = slice_by_index(begin = var_8250_begin_0, end = var_8250_end_0, end_mask = var_8250_end_mask_0, x = transpose_12)[name = tensor("op_8250_cast_fp16")]; tensor var_8254_begin_0 = const()[name = tensor("op_8254_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_8254_end_0 = const()[name = tensor("op_8254_end_0"), val = tensor([2, 77, 1, 384])]; tensor var_8254_end_mask_0 = const()[name = tensor("op_8254_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8254_cast_fp16 = slice_by_index(begin = var_8254_begin_0, end = var_8254_end_0, end_mask = var_8254_end_mask_0, x = transpose_12)[name = tensor("op_8254_cast_fp16")]; tensor var_8258_begin_0 = const()[name = tensor("op_8258_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_8258_end_0 = const()[name = tensor("op_8258_end_0"), val = tensor([2, 77, 1, 448])]; tensor var_8258_end_mask_0 = const()[name = tensor("op_8258_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8258_cast_fp16 = slice_by_index(begin = var_8258_begin_0, end = var_8258_end_0, end_mask = var_8258_end_mask_0, x = transpose_12)[name = tensor("op_8258_cast_fp16")]; tensor var_8262_begin_0 = const()[name = tensor("op_8262_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_8262_end_0 = const()[name = tensor("op_8262_end_0"), val = tensor([2, 77, 1, 512])]; tensor var_8262_end_mask_0 = const()[name = tensor("op_8262_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8262_cast_fp16 = slice_by_index(begin = var_8262_begin_0, end = var_8262_end_0, end_mask = var_8262_end_mask_0, x = transpose_12)[name = tensor("op_8262_cast_fp16")]; tensor var_8266_begin_0 = const()[name = tensor("op_8266_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8266_end_0 = const()[name = tensor("op_8266_end_0"), val = tensor([2, 77, 1, 576])]; tensor var_8266_end_mask_0 = const()[name = tensor("op_8266_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8266_cast_fp16 = slice_by_index(begin = var_8266_begin_0, end = var_8266_end_0, end_mask = var_8266_end_mask_0, x = transpose_12)[name = tensor("op_8266_cast_fp16")]; tensor var_8270_begin_0 = const()[name = tensor("op_8270_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_8270_end_0 = const()[name = tensor("op_8270_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_8270_end_mask_0 = const()[name = tensor("op_8270_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8270_cast_fp16 = slice_by_index(begin = var_8270_begin_0, end = var_8270_end_0, end_mask = var_8270_end_mask_0, x = transpose_12)[name = tensor("op_8270_cast_fp16")]; tensor var_8274_begin_0 = const()[name = tensor("op_8274_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_8274_end_0 = const()[name = tensor("op_8274_end_0"), val = tensor([2, 77, 1, 704])]; tensor var_8274_end_mask_0 = const()[name = tensor("op_8274_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8274_cast_fp16 = slice_by_index(begin = var_8274_begin_0, end = var_8274_end_0, end_mask = var_8274_end_mask_0, x = transpose_12)[name = tensor("op_8274_cast_fp16")]; tensor var_8278_begin_0 = const()[name = tensor("op_8278_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_8278_end_0 = const()[name = tensor("op_8278_end_0"), val = tensor([2, 77, 1, 768])]; tensor var_8278_end_mask_0 = const()[name = tensor("op_8278_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8278_cast_fp16 = slice_by_index(begin = var_8278_begin_0, end = var_8278_end_0, end_mask = var_8278_end_mask_0, x = transpose_12)[name = tensor("op_8278_cast_fp16")]; tensor var_8282_begin_0 = const()[name = tensor("op_8282_begin_0"), val = tensor([0, 0, 0, 768])]; tensor var_8282_end_0 = const()[name = tensor("op_8282_end_0"), val = tensor([2, 77, 1, 832])]; tensor var_8282_end_mask_0 = const()[name = tensor("op_8282_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8282_cast_fp16 = slice_by_index(begin = var_8282_begin_0, end = var_8282_end_0, end_mask = var_8282_end_mask_0, x = transpose_12)[name = tensor("op_8282_cast_fp16")]; tensor var_8286_begin_0 = const()[name = tensor("op_8286_begin_0"), val = tensor([0, 0, 0, 832])]; tensor var_8286_end_0 = const()[name = tensor("op_8286_end_0"), val = tensor([2, 77, 1, 896])]; tensor var_8286_end_mask_0 = const()[name = tensor("op_8286_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8286_cast_fp16 = slice_by_index(begin = var_8286_begin_0, end = var_8286_end_0, end_mask = var_8286_end_mask_0, x = transpose_12)[name = tensor("op_8286_cast_fp16")]; tensor var_8290_begin_0 = const()[name = tensor("op_8290_begin_0"), val = tensor([0, 0, 0, 896])]; tensor var_8290_end_0 = const()[name = tensor("op_8290_end_0"), val = tensor([2, 77, 1, 960])]; tensor var_8290_end_mask_0 = const()[name = tensor("op_8290_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8290_cast_fp16 = slice_by_index(begin = var_8290_begin_0, end = var_8290_end_0, end_mask = var_8290_end_mask_0, x = transpose_12)[name = tensor("op_8290_cast_fp16")]; tensor var_8294_begin_0 = const()[name = tensor("op_8294_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_8294_end_0 = const()[name = tensor("op_8294_end_0"), val = tensor([2, 77, 1, 1024])]; tensor var_8294_end_mask_0 = const()[name = tensor("op_8294_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8294_cast_fp16 = slice_by_index(begin = var_8294_begin_0, end = var_8294_end_0, end_mask = var_8294_end_mask_0, x = transpose_12)[name = tensor("op_8294_cast_fp16")]; tensor var_8298_begin_0 = const()[name = tensor("op_8298_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_8298_end_0 = const()[name = tensor("op_8298_end_0"), val = tensor([2, 77, 1, 1088])]; tensor var_8298_end_mask_0 = const()[name = tensor("op_8298_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8298_cast_fp16 = slice_by_index(begin = var_8298_begin_0, end = var_8298_end_0, end_mask = var_8298_end_mask_0, x = transpose_12)[name = tensor("op_8298_cast_fp16")]; tensor var_8302_begin_0 = const()[name = tensor("op_8302_begin_0"), val = tensor([0, 0, 0, 1088])]; tensor var_8302_end_0 = const()[name = tensor("op_8302_end_0"), val = tensor([2, 77, 1, 1152])]; tensor var_8302_end_mask_0 = const()[name = tensor("op_8302_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8302_cast_fp16 = slice_by_index(begin = var_8302_begin_0, end = var_8302_end_0, end_mask = var_8302_end_mask_0, x = transpose_12)[name = tensor("op_8302_cast_fp16")]; tensor var_8306_begin_0 = const()[name = tensor("op_8306_begin_0"), val = tensor([0, 0, 0, 1152])]; tensor var_8306_end_0 = const()[name = tensor("op_8306_end_0"), val = tensor([2, 77, 1, 1216])]; tensor var_8306_end_mask_0 = const()[name = tensor("op_8306_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8306_cast_fp16 = slice_by_index(begin = var_8306_begin_0, end = var_8306_end_0, end_mask = var_8306_end_mask_0, x = transpose_12)[name = tensor("op_8306_cast_fp16")]; tensor var_8310_begin_0 = const()[name = tensor("op_8310_begin_0"), val = tensor([0, 0, 0, 1216])]; tensor var_8310_end_0 = const()[name = tensor("op_8310_end_0"), val = tensor([2, 77, 1, 1280])]; tensor var_8310_end_mask_0 = const()[name = tensor("op_8310_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8310_cast_fp16 = slice_by_index(begin = var_8310_begin_0, end = var_8310_end_0, end_mask = var_8310_end_mask_0, x = transpose_12)[name = tensor("op_8310_cast_fp16")]; tensor var_8312_begin_0 = const()[name = tensor("op_8312_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8312_end_0 = const()[name = tensor("op_8312_end_0"), val = tensor([2, 64, 1, 77])]; tensor var_8312_end_mask_0 = const()[name = tensor("op_8312_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8312_cast_fp16 = slice_by_index(begin = var_8312_begin_0, end = var_8312_end_0, end_mask = var_8312_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8312_cast_fp16")]; tensor var_8316_begin_0 = const()[name = tensor("op_8316_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_8316_end_0 = const()[name = tensor("op_8316_end_0"), val = tensor([2, 128, 1, 77])]; tensor var_8316_end_mask_0 = const()[name = tensor("op_8316_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8316_cast_fp16 = slice_by_index(begin = var_8316_begin_0, end = var_8316_end_0, end_mask = var_8316_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8316_cast_fp16")]; tensor var_8320_begin_0 = const()[name = tensor("op_8320_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_8320_end_0 = const()[name = tensor("op_8320_end_0"), val = tensor([2, 192, 1, 77])]; tensor var_8320_end_mask_0 = const()[name = tensor("op_8320_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8320_cast_fp16 = slice_by_index(begin = var_8320_begin_0, end = var_8320_end_0, end_mask = var_8320_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8320_cast_fp16")]; tensor var_8324_begin_0 = const()[name = tensor("op_8324_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_8324_end_0 = const()[name = tensor("op_8324_end_0"), val = tensor([2, 256, 1, 77])]; tensor var_8324_end_mask_0 = const()[name = tensor("op_8324_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8324_cast_fp16 = slice_by_index(begin = var_8324_begin_0, end = var_8324_end_0, end_mask = var_8324_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8324_cast_fp16")]; tensor var_8328_begin_0 = const()[name = tensor("op_8328_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_8328_end_0 = const()[name = tensor("op_8328_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_8328_end_mask_0 = const()[name = tensor("op_8328_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8328_cast_fp16 = slice_by_index(begin = var_8328_begin_0, end = var_8328_end_0, end_mask = var_8328_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8328_cast_fp16")]; tensor var_8332_begin_0 = const()[name = tensor("op_8332_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8332_end_0 = const()[name = tensor("op_8332_end_0"), val = tensor([2, 384, 1, 77])]; tensor var_8332_end_mask_0 = const()[name = tensor("op_8332_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8332_cast_fp16 = slice_by_index(begin = var_8332_begin_0, end = var_8332_end_0, end_mask = var_8332_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8332_cast_fp16")]; tensor var_8336_begin_0 = const()[name = tensor("op_8336_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_8336_end_0 = const()[name = tensor("op_8336_end_0"), val = tensor([2, 448, 1, 77])]; tensor var_8336_end_mask_0 = const()[name = tensor("op_8336_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8336_cast_fp16 = slice_by_index(begin = var_8336_begin_0, end = var_8336_end_0, end_mask = var_8336_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8336_cast_fp16")]; tensor var_8340_begin_0 = const()[name = tensor("op_8340_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_8340_end_0 = const()[name = tensor("op_8340_end_0"), val = tensor([2, 512, 1, 77])]; tensor var_8340_end_mask_0 = const()[name = tensor("op_8340_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8340_cast_fp16 = slice_by_index(begin = var_8340_begin_0, end = var_8340_end_0, end_mask = var_8340_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8340_cast_fp16")]; tensor var_8344_begin_0 = const()[name = tensor("op_8344_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_8344_end_0 = const()[name = tensor("op_8344_end_0"), val = tensor([2, 576, 1, 77])]; tensor var_8344_end_mask_0 = const()[name = tensor("op_8344_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8344_cast_fp16 = slice_by_index(begin = var_8344_begin_0, end = var_8344_end_0, end_mask = var_8344_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8344_cast_fp16")]; tensor var_8348_begin_0 = const()[name = tensor("op_8348_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_8348_end_0 = const()[name = tensor("op_8348_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_8348_end_mask_0 = const()[name = tensor("op_8348_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8348_cast_fp16 = slice_by_index(begin = var_8348_begin_0, end = var_8348_end_0, end_mask = var_8348_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8348_cast_fp16")]; tensor var_8352_begin_0 = const()[name = tensor("op_8352_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_8352_end_0 = const()[name = tensor("op_8352_end_0"), val = tensor([2, 704, 1, 77])]; tensor var_8352_end_mask_0 = const()[name = tensor("op_8352_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8352_cast_fp16 = slice_by_index(begin = var_8352_begin_0, end = var_8352_end_0, end_mask = var_8352_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8352_cast_fp16")]; tensor var_8356_begin_0 = const()[name = tensor("op_8356_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_8356_end_0 = const()[name = tensor("op_8356_end_0"), val = tensor([2, 768, 1, 77])]; tensor var_8356_end_mask_0 = const()[name = tensor("op_8356_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8356_cast_fp16 = slice_by_index(begin = var_8356_begin_0, end = var_8356_end_0, end_mask = var_8356_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8356_cast_fp16")]; tensor var_8360_begin_0 = const()[name = tensor("op_8360_begin_0"), val = tensor([0, 768, 0, 0])]; tensor var_8360_end_0 = const()[name = tensor("op_8360_end_0"), val = tensor([2, 832, 1, 77])]; tensor var_8360_end_mask_0 = const()[name = tensor("op_8360_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8360_cast_fp16 = slice_by_index(begin = var_8360_begin_0, end = var_8360_end_0, end_mask = var_8360_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8360_cast_fp16")]; tensor var_8364_begin_0 = const()[name = tensor("op_8364_begin_0"), val = tensor([0, 832, 0, 0])]; tensor var_8364_end_0 = const()[name = tensor("op_8364_end_0"), val = tensor([2, 896, 1, 77])]; tensor var_8364_end_mask_0 = const()[name = tensor("op_8364_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8364_cast_fp16 = slice_by_index(begin = var_8364_begin_0, end = var_8364_end_0, end_mask = var_8364_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8364_cast_fp16")]; tensor var_8368_begin_0 = const()[name = tensor("op_8368_begin_0"), val = tensor([0, 896, 0, 0])]; tensor var_8368_end_0 = const()[name = tensor("op_8368_end_0"), val = tensor([2, 960, 1, 77])]; tensor var_8368_end_mask_0 = const()[name = tensor("op_8368_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8368_cast_fp16 = slice_by_index(begin = var_8368_begin_0, end = var_8368_end_0, end_mask = var_8368_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8368_cast_fp16")]; tensor var_8372_begin_0 = const()[name = tensor("op_8372_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_8372_end_0 = const()[name = tensor("op_8372_end_0"), val = tensor([2, 1024, 1, 77])]; tensor var_8372_end_mask_0 = const()[name = tensor("op_8372_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8372_cast_fp16 = slice_by_index(begin = var_8372_begin_0, end = var_8372_end_0, end_mask = var_8372_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8372_cast_fp16")]; tensor var_8376_begin_0 = const()[name = tensor("op_8376_begin_0"), val = tensor([0, 1024, 0, 0])]; tensor var_8376_end_0 = const()[name = tensor("op_8376_end_0"), val = tensor([2, 1088, 1, 77])]; tensor var_8376_end_mask_0 = const()[name = tensor("op_8376_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8376_cast_fp16 = slice_by_index(begin = var_8376_begin_0, end = var_8376_end_0, end_mask = var_8376_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8376_cast_fp16")]; tensor var_8380_begin_0 = const()[name = tensor("op_8380_begin_0"), val = tensor([0, 1088, 0, 0])]; tensor var_8380_end_0 = const()[name = tensor("op_8380_end_0"), val = tensor([2, 1152, 1, 77])]; tensor var_8380_end_mask_0 = const()[name = tensor("op_8380_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8380_cast_fp16 = slice_by_index(begin = var_8380_begin_0, end = var_8380_end_0, end_mask = var_8380_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8380_cast_fp16")]; tensor var_8384_begin_0 = const()[name = tensor("op_8384_begin_0"), val = tensor([0, 1152, 0, 0])]; tensor var_8384_end_0 = const()[name = tensor("op_8384_end_0"), val = tensor([2, 1216, 1, 77])]; tensor var_8384_end_mask_0 = const()[name = tensor("op_8384_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8384_cast_fp16 = slice_by_index(begin = var_8384_begin_0, end = var_8384_end_0, end_mask = var_8384_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8384_cast_fp16")]; tensor var_8388_begin_0 = const()[name = tensor("op_8388_begin_0"), val = tensor([0, 1216, 0, 0])]; tensor var_8388_end_0 = const()[name = tensor("op_8388_end_0"), val = tensor([2, 1280, 1, 77])]; tensor var_8388_end_mask_0 = const()[name = tensor("op_8388_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8388_cast_fp16 = slice_by_index(begin = var_8388_begin_0, end = var_8388_end_0, end_mask = var_8388_end_mask_0, x = v_39_cast_fp16)[name = tensor("op_8388_cast_fp16")]; tensor var_8392_equation_0 = const()[name = tensor("op_8392_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8392_cast_fp16 = einsum(equation = var_8392_equation_0, values = (var_8234_cast_fp16, var_8151_cast_fp16))[name = tensor("op_8392_cast_fp16")]; tensor var_8393_to_fp16 = const()[name = tensor("op_8393_to_fp16"), val = tensor(0x1p-3)]; tensor aw_561_cast_fp16 = mul(x = var_8392_cast_fp16, y = var_8393_to_fp16)[name = tensor("aw_561_cast_fp16")]; tensor var_8396_equation_0 = const()[name = tensor("op_8396_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8396_cast_fp16 = einsum(equation = var_8396_equation_0, values = (var_8238_cast_fp16, var_8155_cast_fp16))[name = tensor("op_8396_cast_fp16")]; tensor var_8397_to_fp16 = const()[name = tensor("op_8397_to_fp16"), val = tensor(0x1p-3)]; tensor aw_563_cast_fp16 = mul(x = var_8396_cast_fp16, y = var_8397_to_fp16)[name = tensor("aw_563_cast_fp16")]; tensor var_8400_equation_0 = const()[name = tensor("op_8400_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8400_cast_fp16 = einsum(equation = var_8400_equation_0, values = (var_8242_cast_fp16, var_8159_cast_fp16))[name = tensor("op_8400_cast_fp16")]; tensor var_8401_to_fp16 = const()[name = tensor("op_8401_to_fp16"), val = tensor(0x1p-3)]; tensor aw_565_cast_fp16 = mul(x = var_8400_cast_fp16, y = var_8401_to_fp16)[name = tensor("aw_565_cast_fp16")]; tensor var_8404_equation_0 = const()[name = tensor("op_8404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8404_cast_fp16 = einsum(equation = var_8404_equation_0, values = (var_8246_cast_fp16, var_8163_cast_fp16))[name = tensor("op_8404_cast_fp16")]; tensor var_8405_to_fp16 = const()[name = tensor("op_8405_to_fp16"), val = tensor(0x1p-3)]; tensor aw_567_cast_fp16 = mul(x = var_8404_cast_fp16, y = var_8405_to_fp16)[name = tensor("aw_567_cast_fp16")]; tensor var_8408_equation_0 = const()[name = tensor("op_8408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8408_cast_fp16 = einsum(equation = var_8408_equation_0, values = (var_8250_cast_fp16, var_8167_cast_fp16))[name = tensor("op_8408_cast_fp16")]; tensor var_8409_to_fp16 = const()[name = tensor("op_8409_to_fp16"), val = tensor(0x1p-3)]; tensor aw_569_cast_fp16 = mul(x = var_8408_cast_fp16, y = var_8409_to_fp16)[name = tensor("aw_569_cast_fp16")]; tensor var_8412_equation_0 = const()[name = tensor("op_8412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8412_cast_fp16 = einsum(equation = var_8412_equation_0, values = (var_8254_cast_fp16, var_8171_cast_fp16))[name = tensor("op_8412_cast_fp16")]; tensor var_8413_to_fp16 = const()[name = tensor("op_8413_to_fp16"), val = tensor(0x1p-3)]; tensor aw_571_cast_fp16 = mul(x = var_8412_cast_fp16, y = var_8413_to_fp16)[name = tensor("aw_571_cast_fp16")]; tensor var_8416_equation_0 = const()[name = tensor("op_8416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8416_cast_fp16 = einsum(equation = var_8416_equation_0, values = (var_8258_cast_fp16, var_8175_cast_fp16))[name = tensor("op_8416_cast_fp16")]; tensor var_8417_to_fp16 = const()[name = tensor("op_8417_to_fp16"), val = tensor(0x1p-3)]; tensor aw_573_cast_fp16 = mul(x = var_8416_cast_fp16, y = var_8417_to_fp16)[name = tensor("aw_573_cast_fp16")]; tensor var_8420_equation_0 = const()[name = tensor("op_8420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8420_cast_fp16 = einsum(equation = var_8420_equation_0, values = (var_8262_cast_fp16, var_8179_cast_fp16))[name = tensor("op_8420_cast_fp16")]; tensor var_8421_to_fp16 = const()[name = tensor("op_8421_to_fp16"), val = tensor(0x1p-3)]; tensor aw_575_cast_fp16 = mul(x = var_8420_cast_fp16, y = var_8421_to_fp16)[name = tensor("aw_575_cast_fp16")]; tensor var_8424_equation_0 = const()[name = tensor("op_8424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8424_cast_fp16 = einsum(equation = var_8424_equation_0, values = (var_8266_cast_fp16, var_8183_cast_fp16))[name = tensor("op_8424_cast_fp16")]; tensor var_8425_to_fp16 = const()[name = tensor("op_8425_to_fp16"), val = tensor(0x1p-3)]; tensor aw_577_cast_fp16 = mul(x = var_8424_cast_fp16, y = var_8425_to_fp16)[name = tensor("aw_577_cast_fp16")]; tensor var_8428_equation_0 = const()[name = tensor("op_8428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8428_cast_fp16 = einsum(equation = var_8428_equation_0, values = (var_8270_cast_fp16, var_8187_cast_fp16))[name = tensor("op_8428_cast_fp16")]; tensor var_8429_to_fp16 = const()[name = tensor("op_8429_to_fp16"), val = tensor(0x1p-3)]; tensor aw_579_cast_fp16 = mul(x = var_8428_cast_fp16, y = var_8429_to_fp16)[name = tensor("aw_579_cast_fp16")]; tensor var_8432_equation_0 = const()[name = tensor("op_8432_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8432_cast_fp16 = einsum(equation = var_8432_equation_0, values = (var_8274_cast_fp16, var_8191_cast_fp16))[name = tensor("op_8432_cast_fp16")]; tensor var_8433_to_fp16 = const()[name = tensor("op_8433_to_fp16"), val = tensor(0x1p-3)]; tensor aw_581_cast_fp16 = mul(x = var_8432_cast_fp16, y = var_8433_to_fp16)[name = tensor("aw_581_cast_fp16")]; tensor var_8436_equation_0 = const()[name = tensor("op_8436_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8436_cast_fp16 = einsum(equation = var_8436_equation_0, values = (var_8278_cast_fp16, var_8195_cast_fp16))[name = tensor("op_8436_cast_fp16")]; tensor var_8437_to_fp16 = const()[name = tensor("op_8437_to_fp16"), val = tensor(0x1p-3)]; tensor aw_583_cast_fp16 = mul(x = var_8436_cast_fp16, y = var_8437_to_fp16)[name = tensor("aw_583_cast_fp16")]; tensor var_8440_equation_0 = const()[name = tensor("op_8440_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8440_cast_fp16 = einsum(equation = var_8440_equation_0, values = (var_8282_cast_fp16, var_8199_cast_fp16))[name = tensor("op_8440_cast_fp16")]; tensor var_8441_to_fp16 = const()[name = tensor("op_8441_to_fp16"), val = tensor(0x1p-3)]; tensor aw_585_cast_fp16 = mul(x = var_8440_cast_fp16, y = var_8441_to_fp16)[name = tensor("aw_585_cast_fp16")]; tensor var_8444_equation_0 = const()[name = tensor("op_8444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8444_cast_fp16 = einsum(equation = var_8444_equation_0, values = (var_8286_cast_fp16, var_8203_cast_fp16))[name = tensor("op_8444_cast_fp16")]; tensor var_8445_to_fp16 = const()[name = tensor("op_8445_to_fp16"), val = tensor(0x1p-3)]; tensor aw_587_cast_fp16 = mul(x = var_8444_cast_fp16, y = var_8445_to_fp16)[name = tensor("aw_587_cast_fp16")]; tensor var_8448_equation_0 = const()[name = tensor("op_8448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8448_cast_fp16 = einsum(equation = var_8448_equation_0, values = (var_8290_cast_fp16, var_8207_cast_fp16))[name = tensor("op_8448_cast_fp16")]; tensor var_8449_to_fp16 = const()[name = tensor("op_8449_to_fp16"), val = tensor(0x1p-3)]; tensor aw_589_cast_fp16 = mul(x = var_8448_cast_fp16, y = var_8449_to_fp16)[name = tensor("aw_589_cast_fp16")]; tensor var_8452_equation_0 = const()[name = tensor("op_8452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8452_cast_fp16 = einsum(equation = var_8452_equation_0, values = (var_8294_cast_fp16, var_8211_cast_fp16))[name = tensor("op_8452_cast_fp16")]; tensor var_8453_to_fp16 = const()[name = tensor("op_8453_to_fp16"), val = tensor(0x1p-3)]; tensor aw_591_cast_fp16 = mul(x = var_8452_cast_fp16, y = var_8453_to_fp16)[name = tensor("aw_591_cast_fp16")]; tensor var_8456_equation_0 = const()[name = tensor("op_8456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8456_cast_fp16 = einsum(equation = var_8456_equation_0, values = (var_8298_cast_fp16, var_8215_cast_fp16))[name = tensor("op_8456_cast_fp16")]; tensor var_8457_to_fp16 = const()[name = tensor("op_8457_to_fp16"), val = tensor(0x1p-3)]; tensor aw_593_cast_fp16 = mul(x = var_8456_cast_fp16, y = var_8457_to_fp16)[name = tensor("aw_593_cast_fp16")]; tensor var_8460_equation_0 = const()[name = tensor("op_8460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8460_cast_fp16 = einsum(equation = var_8460_equation_0, values = (var_8302_cast_fp16, var_8219_cast_fp16))[name = tensor("op_8460_cast_fp16")]; tensor var_8461_to_fp16 = const()[name = tensor("op_8461_to_fp16"), val = tensor(0x1p-3)]; tensor aw_595_cast_fp16 = mul(x = var_8460_cast_fp16, y = var_8461_to_fp16)[name = tensor("aw_595_cast_fp16")]; tensor var_8464_equation_0 = const()[name = tensor("op_8464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8464_cast_fp16 = einsum(equation = var_8464_equation_0, values = (var_8306_cast_fp16, var_8223_cast_fp16))[name = tensor("op_8464_cast_fp16")]; tensor var_8465_to_fp16 = const()[name = tensor("op_8465_to_fp16"), val = tensor(0x1p-3)]; tensor aw_597_cast_fp16 = mul(x = var_8464_cast_fp16, y = var_8465_to_fp16)[name = tensor("aw_597_cast_fp16")]; tensor var_8468_equation_0 = const()[name = tensor("op_8468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8468_cast_fp16 = einsum(equation = var_8468_equation_0, values = (var_8310_cast_fp16, var_8227_cast_fp16))[name = tensor("op_8468_cast_fp16")]; tensor var_8469_to_fp16 = const()[name = tensor("op_8469_to_fp16"), val = tensor(0x1p-3)]; tensor aw_599_cast_fp16 = mul(x = var_8468_cast_fp16, y = var_8469_to_fp16)[name = tensor("aw_599_cast_fp16")]; tensor var_8471_cast_fp16 = softmax(axis = var_5599, x = aw_561_cast_fp16)[name = tensor("op_8471_cast_fp16")]; tensor var_8472_cast_fp16 = softmax(axis = var_5599, x = aw_563_cast_fp16)[name = tensor("op_8472_cast_fp16")]; tensor var_8473_cast_fp16 = softmax(axis = var_5599, x = aw_565_cast_fp16)[name = tensor("op_8473_cast_fp16")]; tensor var_8474_cast_fp16 = softmax(axis = var_5599, x = aw_567_cast_fp16)[name = tensor("op_8474_cast_fp16")]; tensor var_8475_cast_fp16 = softmax(axis = var_5599, x = aw_569_cast_fp16)[name = tensor("op_8475_cast_fp16")]; tensor var_8476_cast_fp16 = softmax(axis = var_5599, x = aw_571_cast_fp16)[name = tensor("op_8476_cast_fp16")]; tensor var_8477_cast_fp16 = softmax(axis = var_5599, x = aw_573_cast_fp16)[name = tensor("op_8477_cast_fp16")]; tensor var_8478_cast_fp16 = softmax(axis = var_5599, x = aw_575_cast_fp16)[name = tensor("op_8478_cast_fp16")]; tensor var_8479_cast_fp16 = softmax(axis = var_5599, x = aw_577_cast_fp16)[name = tensor("op_8479_cast_fp16")]; tensor var_8480_cast_fp16 = softmax(axis = var_5599, x = aw_579_cast_fp16)[name = tensor("op_8480_cast_fp16")]; tensor var_8481_cast_fp16 = softmax(axis = var_5599, x = aw_581_cast_fp16)[name = tensor("op_8481_cast_fp16")]; tensor var_8482_cast_fp16 = softmax(axis = var_5599, x = aw_583_cast_fp16)[name = tensor("op_8482_cast_fp16")]; tensor var_8483_cast_fp16 = softmax(axis = var_5599, x = aw_585_cast_fp16)[name = tensor("op_8483_cast_fp16")]; tensor var_8484_cast_fp16 = softmax(axis = var_5599, x = aw_587_cast_fp16)[name = tensor("op_8484_cast_fp16")]; tensor var_8485_cast_fp16 = softmax(axis = var_5599, x = aw_589_cast_fp16)[name = tensor("op_8485_cast_fp16")]; tensor var_8486_cast_fp16 = softmax(axis = var_5599, x = aw_591_cast_fp16)[name = tensor("op_8486_cast_fp16")]; tensor var_8487_cast_fp16 = softmax(axis = var_5599, x = aw_593_cast_fp16)[name = tensor("op_8487_cast_fp16")]; tensor var_8488_cast_fp16 = softmax(axis = var_5599, x = aw_595_cast_fp16)[name = tensor("op_8488_cast_fp16")]; tensor var_8489_cast_fp16 = softmax(axis = var_5599, x = aw_597_cast_fp16)[name = tensor("op_8489_cast_fp16")]; tensor var_8490_cast_fp16 = softmax(axis = var_5599, x = aw_599_cast_fp16)[name = tensor("op_8490_cast_fp16")]; tensor var_8492_equation_0 = const()[name = tensor("op_8492_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8492_cast_fp16 = einsum(equation = var_8492_equation_0, values = (var_8312_cast_fp16, var_8471_cast_fp16))[name = tensor("op_8492_cast_fp16")]; tensor var_8494_equation_0 = const()[name = tensor("op_8494_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8494_cast_fp16 = einsum(equation = var_8494_equation_0, values = (var_8316_cast_fp16, var_8472_cast_fp16))[name = tensor("op_8494_cast_fp16")]; tensor var_8496_equation_0 = const()[name = tensor("op_8496_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8496_cast_fp16 = einsum(equation = var_8496_equation_0, values = (var_8320_cast_fp16, var_8473_cast_fp16))[name = tensor("op_8496_cast_fp16")]; tensor var_8498_equation_0 = const()[name = tensor("op_8498_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8498_cast_fp16 = einsum(equation = var_8498_equation_0, values = (var_8324_cast_fp16, var_8474_cast_fp16))[name = tensor("op_8498_cast_fp16")]; tensor var_8500_equation_0 = const()[name = tensor("op_8500_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8500_cast_fp16 = einsum(equation = var_8500_equation_0, values = (var_8328_cast_fp16, var_8475_cast_fp16))[name = tensor("op_8500_cast_fp16")]; tensor var_8502_equation_0 = const()[name = tensor("op_8502_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8502_cast_fp16 = einsum(equation = var_8502_equation_0, values = (var_8332_cast_fp16, var_8476_cast_fp16))[name = tensor("op_8502_cast_fp16")]; tensor var_8504_equation_0 = const()[name = tensor("op_8504_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8504_cast_fp16 = einsum(equation = var_8504_equation_0, values = (var_8336_cast_fp16, var_8477_cast_fp16))[name = tensor("op_8504_cast_fp16")]; tensor var_8506_equation_0 = const()[name = tensor("op_8506_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8506_cast_fp16 = einsum(equation = var_8506_equation_0, values = (var_8340_cast_fp16, var_8478_cast_fp16))[name = tensor("op_8506_cast_fp16")]; tensor var_8508_equation_0 = const()[name = tensor("op_8508_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8508_cast_fp16 = einsum(equation = var_8508_equation_0, values = (var_8344_cast_fp16, var_8479_cast_fp16))[name = tensor("op_8508_cast_fp16")]; tensor var_8510_equation_0 = const()[name = tensor("op_8510_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8510_cast_fp16 = einsum(equation = var_8510_equation_0, values = (var_8348_cast_fp16, var_8480_cast_fp16))[name = tensor("op_8510_cast_fp16")]; tensor var_8512_equation_0 = const()[name = tensor("op_8512_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8512_cast_fp16 = einsum(equation = var_8512_equation_0, values = (var_8352_cast_fp16, var_8481_cast_fp16))[name = tensor("op_8512_cast_fp16")]; tensor var_8514_equation_0 = const()[name = tensor("op_8514_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8514_cast_fp16 = einsum(equation = var_8514_equation_0, values = (var_8356_cast_fp16, var_8482_cast_fp16))[name = tensor("op_8514_cast_fp16")]; tensor var_8516_equation_0 = const()[name = tensor("op_8516_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8516_cast_fp16 = einsum(equation = var_8516_equation_0, values = (var_8360_cast_fp16, var_8483_cast_fp16))[name = tensor("op_8516_cast_fp16")]; tensor var_8518_equation_0 = const()[name = tensor("op_8518_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8518_cast_fp16 = einsum(equation = var_8518_equation_0, values = (var_8364_cast_fp16, var_8484_cast_fp16))[name = tensor("op_8518_cast_fp16")]; tensor var_8520_equation_0 = const()[name = tensor("op_8520_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8520_cast_fp16 = einsum(equation = var_8520_equation_0, values = (var_8368_cast_fp16, var_8485_cast_fp16))[name = tensor("op_8520_cast_fp16")]; tensor var_8522_equation_0 = const()[name = tensor("op_8522_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8522_cast_fp16 = einsum(equation = var_8522_equation_0, values = (var_8372_cast_fp16, var_8486_cast_fp16))[name = tensor("op_8522_cast_fp16")]; tensor var_8524_equation_0 = const()[name = tensor("op_8524_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8524_cast_fp16 = einsum(equation = var_8524_equation_0, values = (var_8376_cast_fp16, var_8487_cast_fp16))[name = tensor("op_8524_cast_fp16")]; tensor var_8526_equation_0 = const()[name = tensor("op_8526_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8526_cast_fp16 = einsum(equation = var_8526_equation_0, values = (var_8380_cast_fp16, var_8488_cast_fp16))[name = tensor("op_8526_cast_fp16")]; tensor var_8528_equation_0 = const()[name = tensor("op_8528_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8528_cast_fp16 = einsum(equation = var_8528_equation_0, values = (var_8384_cast_fp16, var_8489_cast_fp16))[name = tensor("op_8528_cast_fp16")]; tensor var_8530_equation_0 = const()[name = tensor("op_8530_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8530_cast_fp16 = einsum(equation = var_8530_equation_0, values = (var_8388_cast_fp16, var_8490_cast_fp16))[name = tensor("op_8530_cast_fp16")]; tensor input_355_interleave_0 = const()[name = tensor("input_355_interleave_0"), val = tensor(false)]; tensor input_355_cast_fp16 = concat(axis = var_5599, interleave = input_355_interleave_0, values = (var_8492_cast_fp16, var_8494_cast_fp16, var_8496_cast_fp16, var_8498_cast_fp16, var_8500_cast_fp16, var_8502_cast_fp16, var_8504_cast_fp16, var_8506_cast_fp16, var_8508_cast_fp16, var_8510_cast_fp16, var_8512_cast_fp16, var_8514_cast_fp16, var_8516_cast_fp16, var_8518_cast_fp16, var_8520_cast_fp16, var_8522_cast_fp16, var_8524_cast_fp16, var_8526_cast_fp16, var_8528_cast_fp16, var_8530_cast_fp16))[name = tensor("input_355_cast_fp16")]; tensor var_8536 = const()[name = tensor("op_8536"), val = tensor([1, 1])]; tensor var_8538 = const()[name = tensor("op_8538"), val = tensor([1, 1])]; tensor var_8540_pad_type_0 = const()[name = tensor("op_8540_pad_type_0"), val = tensor("custom")]; tensor var_8540_pad_0 = const()[name = tensor("op_8540_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(579606464)))]; tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(582883328)))]; tensor var_8540_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_8538, groups = var_5599, pad = var_8540_pad_0, pad_type = var_8540_pad_type_0, strides = var_8536, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_355_cast_fp16)[name = tensor("op_8540_cast_fp16")]; tensor inputs_59_cast_fp16 = add(x = var_8540_cast_fp16, y = inputs_57_cast_fp16)[name = tensor("inputs_59_cast_fp16")]; tensor var_8544 = const()[name = tensor("op_8544"), val = tensor([1])]; tensor channels_mean_59_cast_fp16 = reduce_mean(axes = var_8544, keep_dims = var_5594, x = inputs_59_cast_fp16)[name = tensor("channels_mean_59_cast_fp16")]; tensor zero_mean_59_cast_fp16 = sub(x = inputs_59_cast_fp16, y = channels_mean_59_cast_fp16)[name = tensor("zero_mean_59_cast_fp16")]; tensor zero_mean_sq_59_cast_fp16 = mul(x = zero_mean_59_cast_fp16, y = zero_mean_59_cast_fp16)[name = tensor("zero_mean_sq_59_cast_fp16")]; tensor var_8548 = const()[name = tensor("op_8548"), val = tensor([1])]; tensor var_8549_cast_fp16 = reduce_mean(axes = var_8548, keep_dims = var_5594, x = zero_mean_sq_59_cast_fp16)[name = tensor("op_8549_cast_fp16")]; tensor var_8550_to_fp16 = const()[name = tensor("op_8550_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_8551_cast_fp16 = add(x = var_8549_cast_fp16, y = var_8550_to_fp16)[name = tensor("op_8551_cast_fp16")]; tensor denom_59_epsilon_0_to_fp16 = const()[name = tensor("denom_59_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_59_cast_fp16 = rsqrt(epsilon = denom_59_epsilon_0_to_fp16, x = var_8551_cast_fp16)[name = tensor("denom_59_cast_fp16")]; tensor out_59_cast_fp16 = mul(x = zero_mean_59_cast_fp16, y = denom_59_cast_fp16)[name = tensor("out_59_cast_fp16")]; tensor var_8555_to_fp16 = const()[name = tensor("op_8555_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(582885952)))]; tensor var_8556_cast_fp16 = add(x = out_59_cast_fp16, y = var_8555_to_fp16)[name = tensor("op_8556_cast_fp16")]; tensor var_8558_to_fp16 = const()[name = tensor("op_8558_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(582888576)))]; tensor input_357_cast_fp16 = mul(x = var_8556_cast_fp16, y = var_8558_to_fp16)[name = tensor("input_357_cast_fp16")]; tensor var_8566 = const()[name = tensor("op_8566"), val = tensor([1, 1])]; tensor var_8568 = const()[name = tensor("op_8568"), val = tensor([1, 1])]; tensor var_8570_pad_type_0 = const()[name = tensor("op_8570_pad_type_0"), val = tensor("custom")]; tensor var_8570_pad_0 = const()[name = tensor("op_8570_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(582891200)))]; tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(609105664)))]; tensor var_8570_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_8568, groups = var_5599, pad = var_8570_pad_0, pad_type = var_8570_pad_type_0, strides = var_8566, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_357_cast_fp16)[name = tensor("op_8570_cast_fp16")]; tensor var_8571_split_sizes_0 = const()[name = tensor("op_8571_split_sizes_0"), val = tensor([5120, 5120])]; tensor var_8571_axis_0 = const()[name = tensor("op_8571_axis_0"), val = tensor(1)]; tensor var_8571_cast_fp16_0, tensor var_8571_cast_fp16_1 = split(axis = var_8571_axis_0, split_sizes = var_8571_split_sizes_0, x = var_8570_cast_fp16)[name = tensor("op_8571_cast_fp16")]; tensor var_8573_mode_0 = const()[name = tensor("op_8573_mode_0"), val = tensor("EXACT")]; tensor var_8573_cast_fp16 = gelu(mode = var_8573_mode_0, x = var_8571_cast_fp16_1)[name = tensor("op_8573_cast_fp16")]; tensor input_359_cast_fp16 = mul(x = var_8571_cast_fp16_0, y = var_8573_cast_fp16)[name = tensor("input_359_cast_fp16")]; tensor var_8577 = const()[name = tensor("op_8577"), val = tensor([1, 1])]; tensor var_8579 = const()[name = tensor("op_8579"), val = tensor([1, 1])]; tensor var_8581_pad_type_0 = const()[name = tensor("op_8581_pad_type_0"), val = tensor("custom")]; tensor var_8581_pad_0 = const()[name = tensor("op_8581_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(609126208)))]; tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622233472)))]; tensor var_8581_cast_fp16 = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_8579, groups = var_5599, pad = var_8581_pad_0, pad_type = var_8581_pad_type_0, strides = var_8577, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_359_cast_fp16)[name = tensor("op_8581_cast_fp16")]; tensor hidden_states_213_cast_fp16 = add(x = var_8581_cast_fp16, y = inputs_59_cast_fp16)[name = tensor("hidden_states_213_cast_fp16")]; tensor var_8583 = const()[name = tensor("op_8583"), val = tensor([2, 1280, 12, 12])]; tensor input_361_cast_fp16 = reshape(shape = var_8583, x = hidden_states_213_cast_fp16)[name = tensor("input_361_cast_fp16")]; tensor var_8587 = const()[name = tensor("op_8587"), val = tensor([1, 1])]; tensor var_8589 = const()[name = tensor("op_8589"), val = tensor([1, 1])]; tensor hidden_states_215_pad_type_0 = const()[name = tensor("hidden_states_215_pad_type_0"), val = tensor("custom")]; tensor hidden_states_215_pad_0 = const()[name = tensor("hidden_states_215_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_1_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622236096)))]; tensor up_blocks_1_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625512960)))]; tensor hidden_states_215_cast_fp16 = conv(bias = up_blocks_1_attentions_2_proj_out_bias_to_fp16, dilations = var_8589, groups = var_5599, pad = hidden_states_215_pad_0, pad_type = hidden_states_215_pad_type_0, strides = var_8587, weight = up_blocks_1_attentions_2_proj_out_weight_to_fp16, x = input_361_cast_fp16)[name = tensor("hidden_states_215_cast_fp16")]; tensor input_363_cast_fp16 = add(x = hidden_states_215_cast_fp16, y = hidden_states_203_cast_fp16)[name = tensor("input_363_cast_fp16")]; tensor input_365_scale_factor_height_0 = const()[name = tensor("input_365_scale_factor_height_0"), val = tensor(0x1p+1)]; tensor input_365_scale_factor_width_0 = const()[name = tensor("input_365_scale_factor_width_0"), val = tensor(0x1p+1)]; tensor input_365_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_365_scale_factor_height_0, scale_factor_width = input_365_scale_factor_width_0, x = input_363_cast_fp16)[name = tensor("input_365_cast_fp16")]; tensor var_8598 = const()[name = tensor("op_8598"), val = tensor([1, 1])]; tensor var_8600 = const()[name = tensor("op_8600"), val = tensor([1, 1])]; tensor hidden_states_217_pad_type_0 = const()[name = tensor("hidden_states_217_pad_type_0"), val = tensor("custom")]; tensor hidden_states_217_pad_0 = const()[name = tensor("hidden_states_217_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_1_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor("up_blocks_1_upsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625515584)))]; tensor up_blocks_1_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor("up_blocks_1_upsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(655006848)))]; tensor hidden_states_217_cast_fp16 = conv(bias = up_blocks_1_upsamplers_0_conv_bias_to_fp16, dilations = var_8600, groups = var_5599, pad = hidden_states_217_pad_0, pad_type = hidden_states_217_pad_type_0, strides = var_8598, weight = up_blocks_1_upsamplers_0_conv_weight_to_fp16, x = input_365_cast_fp16)[name = tensor("hidden_states_217_cast_fp16")]; tensor var_8622 = const()[name = tensor("op_8622"), val = tensor(true)]; tensor var_8627 = const()[name = tensor("op_8627"), val = tensor(1)]; tensor input_367_interleave_0 = const()[name = tensor("input_367_interleave_0"), val = tensor(false)]; tensor cast_7 = cast(dtype = cast_6_dtype_0, x = input_115_cast_fp16)[name = tensor("cast_7")]; tensor input_367_cast_fp16 = concat(axis = var_8627, interleave = input_367_interleave_0, values = (hidden_states_217_cast_fp16, cast_7))[name = tensor("input_367_cast_fp16")]; tensor reshape_168_shape_0 = const()[name = tensor("reshape_168_shape_0"), val = tensor([2, 32, 60, 24, 24])]; tensor reshape_168_cast_fp16 = reshape(shape = reshape_168_shape_0, x = input_367_cast_fp16)[name = tensor("reshape_168_cast_fp16")]; tensor reduce_mean_126_axes_0 = const()[name = tensor("reduce_mean_126_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_126_keep_dims_0 = const()[name = tensor("reduce_mean_126_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_126_cast_fp16 = reduce_mean(axes = reduce_mean_126_axes_0, keep_dims = reduce_mean_126_keep_dims_0, x = reshape_168_cast_fp16)[name = tensor("reduce_mean_126_cast_fp16")]; tensor sub_84_cast_fp16 = sub(x = reshape_168_cast_fp16, y = reduce_mean_126_cast_fp16)[name = tensor("sub_84_cast_fp16")]; tensor square_42_cast_fp16 = square(x = sub_84_cast_fp16)[name = tensor("square_42_cast_fp16")]; tensor reduce_mean_128_axes_0 = const()[name = tensor("reduce_mean_128_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_128_keep_dims_0 = const()[name = tensor("reduce_mean_128_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_128_cast_fp16 = reduce_mean(axes = reduce_mean_128_axes_0, keep_dims = reduce_mean_128_keep_dims_0, x = square_42_cast_fp16)[name = tensor("reduce_mean_128_cast_fp16")]; tensor add_84_y_0_to_fp16 = const()[name = tensor("add_84_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_84_cast_fp16 = add(x = reduce_mean_128_cast_fp16, y = add_84_y_0_to_fp16)[name = tensor("add_84_cast_fp16")]; tensor sqrt_42_cast_fp16 = sqrt(x = add_84_cast_fp16)[name = tensor("sqrt_42_cast_fp16")]; tensor real_div_42_cast_fp16 = real_div(x = sub_84_cast_fp16, y = sqrt_42_cast_fp16)[name = tensor("real_div_42_cast_fp16")]; tensor reshape_169_shape_0 = const()[name = tensor("reshape_169_shape_0"), val = tensor([2, 1920, 24, 24])]; tensor reshape_169_cast_fp16 = reshape(shape = reshape_169_shape_0, x = real_div_42_cast_fp16)[name = tensor("reshape_169_cast_fp16")]; tensor add_85_gamma_0_to_fp16 = const()[name = tensor("add_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(655009472)))]; tensor add_85_beta_0_to_fp16 = const()[name = tensor("add_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(655013376)))]; tensor add_85_epsilon_0_to_fp16 = const()[name = tensor("add_85_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_85_cast_fp16 = batch_norm(beta = add_85_beta_0_to_fp16, epsilon = add_85_epsilon_0_to_fp16, gamma = add_85_gamma_0_to_fp16, mean = add_79_mean_0_to_fp16, variance = add_79_variance_0_to_fp16, x = reshape_169_cast_fp16)[name = tensor("add_85_cast_fp16")]; tensor input_371_cast_fp16 = silu(x = add_85_cast_fp16)[name = tensor("input_371_cast_fp16")]; tensor var_8656 = const()[name = tensor("op_8656"), val = tensor([1, 1])]; tensor var_8658 = const()[name = tensor("op_8658"), val = tensor([1, 1])]; tensor hidden_states_219_pad_type_0 = const()[name = tensor("hidden_states_219_pad_type_0"), val = tensor("custom")]; tensor hidden_states_219_pad_0 = const()[name = tensor("hidden_states_219_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_2_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(655017280)))]; tensor up_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(677135744)))]; tensor hidden_states_219_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_8658, groups = var_8627, pad = hidden_states_219_pad_0, pad_type = hidden_states_219_pad_type_0, strides = var_8656, weight = up_blocks_2_resnets_0_conv1_weight_to_fp16, x = input_371_cast_fp16)[name = tensor("hidden_states_219_cast_fp16")]; tensor var_8664 = const()[name = tensor("op_8664"), val = tensor([1, 1])]; tensor var_8666 = const()[name = tensor("op_8666"), val = tensor([1, 1])]; tensor temb_33_pad_type_0 = const()[name = tensor("temb_33_pad_type_0"), val = tensor("custom")]; tensor temb_33_pad_0 = const()[name = tensor("temb_33_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(677137088)))]; tensor up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(678775552)))]; tensor temb_33_cast_fp16 = conv(bias = up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_8666, groups = var_8627, pad = temb_33_pad_0, pad_type = temb_33_pad_type_0, strides = var_8664, weight = up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16, x = cast_12)[name = tensor("temb_33_cast_fp16")]; tensor input_375_cast_fp16 = add(x = hidden_states_219_cast_fp16, y = temb_33_cast_fp16)[name = tensor("input_375_cast_fp16")]; tensor reshape_172_shape_0 = const()[name = tensor("reshape_172_shape_0"), val = tensor([2, 32, 20, 24, 24])]; tensor reshape_172_cast_fp16 = reshape(shape = reshape_172_shape_0, x = input_375_cast_fp16)[name = tensor("reshape_172_cast_fp16")]; tensor reduce_mean_129_axes_0 = const()[name = tensor("reduce_mean_129_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_129_keep_dims_0 = const()[name = tensor("reduce_mean_129_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_129_cast_fp16 = reduce_mean(axes = reduce_mean_129_axes_0, keep_dims = reduce_mean_129_keep_dims_0, x = reshape_172_cast_fp16)[name = tensor("reduce_mean_129_cast_fp16")]; tensor sub_86_cast_fp16 = sub(x = reshape_172_cast_fp16, y = reduce_mean_129_cast_fp16)[name = tensor("sub_86_cast_fp16")]; tensor square_43_cast_fp16 = square(x = sub_86_cast_fp16)[name = tensor("square_43_cast_fp16")]; tensor reduce_mean_131_axes_0 = const()[name = tensor("reduce_mean_131_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_131_keep_dims_0 = const()[name = tensor("reduce_mean_131_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_131_cast_fp16 = reduce_mean(axes = reduce_mean_131_axes_0, keep_dims = reduce_mean_131_keep_dims_0, x = square_43_cast_fp16)[name = tensor("reduce_mean_131_cast_fp16")]; tensor add_86_y_0_to_fp16 = const()[name = tensor("add_86_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_86_cast_fp16 = add(x = reduce_mean_131_cast_fp16, y = add_86_y_0_to_fp16)[name = tensor("add_86_cast_fp16")]; tensor sqrt_43_cast_fp16 = sqrt(x = add_86_cast_fp16)[name = tensor("sqrt_43_cast_fp16")]; tensor real_div_43_cast_fp16 = real_div(x = sub_86_cast_fp16, y = sqrt_43_cast_fp16)[name = tensor("real_div_43_cast_fp16")]; tensor reshape_173_shape_0 = const()[name = tensor("reshape_173_shape_0"), val = tensor([2, 640, 24, 24])]; tensor reshape_173_cast_fp16 = reshape(shape = reshape_173_shape_0, x = real_div_43_cast_fp16)[name = tensor("reshape_173_cast_fp16")]; tensor add_87_gamma_0_to_fp16 = const()[name = tensor("add_87_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(678776896)))]; tensor add_87_beta_0_to_fp16 = const()[name = tensor("add_87_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(678778240)))]; tensor add_87_epsilon_0_to_fp16 = const()[name = tensor("add_87_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_87_cast_fp16 = batch_norm(beta = add_87_beta_0_to_fp16, epsilon = add_87_epsilon_0_to_fp16, gamma = add_87_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_173_cast_fp16)[name = tensor("add_87_cast_fp16")]; tensor input_379_cast_fp16 = silu(x = add_87_cast_fp16)[name = tensor("input_379_cast_fp16")]; tensor var_8676 = const()[name = tensor("op_8676"), val = tensor([1, 1])]; tensor var_8678 = const()[name = tensor("op_8678"), val = tensor([1, 1])]; tensor hidden_states_221_pad_type_0 = const()[name = tensor("hidden_states_221_pad_type_0"), val = tensor("custom")]; tensor hidden_states_221_pad_0 = const()[name = tensor("hidden_states_221_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_2_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(678779584)))]; tensor up_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686152448)))]; tensor hidden_states_221_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_8678, groups = var_8627, pad = hidden_states_221_pad_0, pad_type = hidden_states_221_pad_type_0, strides = var_8676, weight = up_blocks_2_resnets_0_conv2_weight_to_fp16, x = input_379_cast_fp16)[name = tensor("hidden_states_221_cast_fp16")]; tensor var_8683 = const()[name = tensor("op_8683"), val = tensor([1, 1])]; tensor var_8685 = const()[name = tensor("op_8685"), val = tensor([1, 1])]; tensor x_17_pad_type_0 = const()[name = tensor("x_17_pad_type_0"), val = tensor("custom")]; tensor x_17_pad_0 = const()[name = tensor("x_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686153792)))]; tensor up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(688611456)))]; tensor x_17_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_8685, groups = var_8627, pad = x_17_pad_0, pad_type = x_17_pad_type_0, strides = var_8683, weight = up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16, x = input_367_cast_fp16)[name = tensor("x_17_cast_fp16")]; tensor hidden_states_223_cast_fp16 = add(x = x_17_cast_fp16, y = hidden_states_221_cast_fp16)[name = tensor("hidden_states_223_cast_fp16")]; tensor reshape_176_shape_0 = const()[name = tensor("reshape_176_shape_0"), val = tensor([2, 32, 20, 24, 24])]; tensor reshape_176_cast_fp16 = reshape(shape = reshape_176_shape_0, x = hidden_states_223_cast_fp16)[name = tensor("reshape_176_cast_fp16")]; tensor reduce_mean_132_axes_0 = const()[name = tensor("reduce_mean_132_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_132_keep_dims_0 = const()[name = tensor("reduce_mean_132_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_132_cast_fp16 = reduce_mean(axes = reduce_mean_132_axes_0, keep_dims = reduce_mean_132_keep_dims_0, x = reshape_176_cast_fp16)[name = tensor("reduce_mean_132_cast_fp16")]; tensor sub_88_cast_fp16 = sub(x = reshape_176_cast_fp16, y = reduce_mean_132_cast_fp16)[name = tensor("sub_88_cast_fp16")]; tensor square_44_cast_fp16 = square(x = sub_88_cast_fp16)[name = tensor("square_44_cast_fp16")]; tensor reduce_mean_134_axes_0 = const()[name = tensor("reduce_mean_134_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_134_keep_dims_0 = const()[name = tensor("reduce_mean_134_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_134_cast_fp16 = reduce_mean(axes = reduce_mean_134_axes_0, keep_dims = reduce_mean_134_keep_dims_0, x = square_44_cast_fp16)[name = tensor("reduce_mean_134_cast_fp16")]; tensor add_88_y_0_to_fp16 = const()[name = tensor("add_88_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_88_cast_fp16 = add(x = reduce_mean_134_cast_fp16, y = add_88_y_0_to_fp16)[name = tensor("add_88_cast_fp16")]; tensor sqrt_44_cast_fp16 = sqrt(x = add_88_cast_fp16)[name = tensor("sqrt_44_cast_fp16")]; tensor real_div_44_cast_fp16 = real_div(x = sub_88_cast_fp16, y = sqrt_44_cast_fp16)[name = tensor("real_div_44_cast_fp16")]; tensor reshape_177_shape_0 = const()[name = tensor("reshape_177_shape_0"), val = tensor([2, 640, 24, 24])]; tensor reshape_177_cast_fp16 = reshape(shape = reshape_177_shape_0, x = real_div_44_cast_fp16)[name = tensor("reshape_177_cast_fp16")]; tensor add_89_gamma_0_to_fp16 = const()[name = tensor("add_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(688612800)))]; tensor add_89_beta_0_to_fp16 = const()[name = tensor("add_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(688614144)))]; tensor add_89_epsilon_0_to_fp16 = const()[name = tensor("add_89_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_89_cast_fp16 = batch_norm(beta = add_89_beta_0_to_fp16, epsilon = add_89_epsilon_0_to_fp16, gamma = add_89_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_177_cast_fp16)[name = tensor("add_89_cast_fp16")]; tensor var_8705 = const()[name = tensor("op_8705"), val = tensor([1, 1])]; tensor var_8707 = const()[name = tensor("op_8707"), val = tensor([1, 1])]; tensor hidden_states_225_pad_type_0 = const()[name = tensor("hidden_states_225_pad_type_0"), val = tensor("custom")]; tensor hidden_states_225_pad_0 = const()[name = tensor("hidden_states_225_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(688615488)))]; tensor up_blocks_2_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689434752)))]; tensor hidden_states_225_cast_fp16 = conv(bias = up_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_8707, groups = var_8627, pad = hidden_states_225_pad_0, pad_type = hidden_states_225_pad_type_0, strides = var_8705, weight = up_blocks_2_attentions_0_proj_in_weight_to_fp16, x = add_89_cast_fp16)[name = tensor("hidden_states_225_cast_fp16")]; tensor var_8712 = const()[name = tensor("op_8712"), val = tensor([2, 640, 1, 576])]; tensor inputs_61_cast_fp16 = reshape(shape = var_8712, x = hidden_states_225_cast_fp16)[name = tensor("inputs_61_cast_fp16")]; tensor var_8722 = const()[name = tensor("op_8722"), val = tensor([1])]; tensor channels_mean_61_cast_fp16 = reduce_mean(axes = var_8722, keep_dims = var_8622, x = inputs_61_cast_fp16)[name = tensor("channels_mean_61_cast_fp16")]; tensor zero_mean_61_cast_fp16 = sub(x = inputs_61_cast_fp16, y = channels_mean_61_cast_fp16)[name = tensor("zero_mean_61_cast_fp16")]; tensor zero_mean_sq_61_cast_fp16 = mul(x = zero_mean_61_cast_fp16, y = zero_mean_61_cast_fp16)[name = tensor("zero_mean_sq_61_cast_fp16")]; tensor var_8726 = const()[name = tensor("op_8726"), val = tensor([1])]; tensor var_8727_cast_fp16 = reduce_mean(axes = var_8726, keep_dims = var_8622, x = zero_mean_sq_61_cast_fp16)[name = tensor("op_8727_cast_fp16")]; tensor var_8728_to_fp16 = const()[name = tensor("op_8728_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_8729_cast_fp16 = add(x = var_8727_cast_fp16, y = var_8728_to_fp16)[name = tensor("op_8729_cast_fp16")]; tensor denom_61_epsilon_0_to_fp16 = const()[name = tensor("denom_61_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_61_cast_fp16 = rsqrt(epsilon = denom_61_epsilon_0_to_fp16, x = var_8729_cast_fp16)[name = tensor("denom_61_cast_fp16")]; tensor out_61_cast_fp16 = mul(x = zero_mean_61_cast_fp16, y = denom_61_cast_fp16)[name = tensor("out_61_cast_fp16")]; tensor var_8733_to_fp16 = const()[name = tensor("op_8733_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689436096)))]; tensor var_8734_cast_fp16 = add(x = out_61_cast_fp16, y = var_8733_to_fp16)[name = tensor("op_8734_cast_fp16")]; tensor var_8736_to_fp16 = const()[name = tensor("op_8736_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689437440)))]; tensor hidden_states_227_cast_fp16 = mul(x = var_8734_cast_fp16, y = var_8736_to_fp16)[name = tensor("hidden_states_227_cast_fp16")]; tensor var_8743 = const()[name = tensor("op_8743"), val = tensor([1, 1])]; tensor var_8745 = const()[name = tensor("op_8745"), val = tensor([1, 1])]; tensor q_41_pad_type_0 = const()[name = tensor("q_41_pad_type_0"), val = tensor("custom")]; tensor q_41_pad_0 = const()[name = tensor("q_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689438784)))]; tensor q_41_cast_fp16 = conv(dilations = var_8745, groups = var_8627, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = var_8743, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_227_cast_fp16)[name = tensor("q_41_cast_fp16")]; tensor var_8749 = const()[name = tensor("op_8749"), val = tensor([1, 1])]; tensor var_8751 = const()[name = tensor("op_8751"), val = tensor([1, 1])]; tensor k_81_pad_type_0 = const()[name = tensor("k_81_pad_type_0"), val = tensor("custom")]; tensor k_81_pad_0 = const()[name = tensor("k_81_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(690258048)))]; tensor k_81_cast_fp16 = conv(dilations = var_8751, groups = var_8627, pad = k_81_pad_0, pad_type = k_81_pad_type_0, strides = var_8749, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_227_cast_fp16)[name = tensor("k_81_cast_fp16")]; tensor var_8755 = const()[name = tensor("op_8755"), val = tensor([1, 1])]; tensor var_8757 = const()[name = tensor("op_8757"), val = tensor([1, 1])]; tensor v_41_pad_type_0 = const()[name = tensor("v_41_pad_type_0"), val = tensor("custom")]; tensor v_41_pad_0 = const()[name = tensor("v_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(691077312)))]; tensor v_41_cast_fp16 = conv(dilations = var_8757, groups = var_8627, pad = v_41_pad_0, pad_type = v_41_pad_type_0, strides = var_8755, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_227_cast_fp16)[name = tensor("v_41_cast_fp16")]; tensor var_8761_begin_0 = const()[name = tensor("op_8761_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8761_end_0 = const()[name = tensor("op_8761_end_0"), val = tensor([2, 64, 1, 576])]; tensor var_8761_end_mask_0 = const()[name = tensor("op_8761_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8761_cast_fp16 = slice_by_index(begin = var_8761_begin_0, end = var_8761_end_0, end_mask = var_8761_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8761_cast_fp16")]; tensor var_8765_begin_0 = const()[name = tensor("op_8765_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_8765_end_0 = const()[name = tensor("op_8765_end_0"), val = tensor([2, 128, 1, 576])]; tensor var_8765_end_mask_0 = const()[name = tensor("op_8765_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8765_cast_fp16 = slice_by_index(begin = var_8765_begin_0, end = var_8765_end_0, end_mask = var_8765_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8765_cast_fp16")]; tensor var_8769_begin_0 = const()[name = tensor("op_8769_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_8769_end_0 = const()[name = tensor("op_8769_end_0"), val = tensor([2, 192, 1, 576])]; tensor var_8769_end_mask_0 = const()[name = tensor("op_8769_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8769_cast_fp16 = slice_by_index(begin = var_8769_begin_0, end = var_8769_end_0, end_mask = var_8769_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8769_cast_fp16")]; tensor var_8773_begin_0 = const()[name = tensor("op_8773_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_8773_end_0 = const()[name = tensor("op_8773_end_0"), val = tensor([2, 256, 1, 576])]; tensor var_8773_end_mask_0 = const()[name = tensor("op_8773_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8773_cast_fp16 = slice_by_index(begin = var_8773_begin_0, end = var_8773_end_0, end_mask = var_8773_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8773_cast_fp16")]; tensor var_8777_begin_0 = const()[name = tensor("op_8777_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_8777_end_0 = const()[name = tensor("op_8777_end_0"), val = tensor([2, 320, 1, 576])]; tensor var_8777_end_mask_0 = const()[name = tensor("op_8777_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8777_cast_fp16 = slice_by_index(begin = var_8777_begin_0, end = var_8777_end_0, end_mask = var_8777_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8777_cast_fp16")]; tensor var_8781_begin_0 = const()[name = tensor("op_8781_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8781_end_0 = const()[name = tensor("op_8781_end_0"), val = tensor([2, 384, 1, 576])]; tensor var_8781_end_mask_0 = const()[name = tensor("op_8781_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8781_cast_fp16 = slice_by_index(begin = var_8781_begin_0, end = var_8781_end_0, end_mask = var_8781_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8781_cast_fp16")]; tensor var_8785_begin_0 = const()[name = tensor("op_8785_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_8785_end_0 = const()[name = tensor("op_8785_end_0"), val = tensor([2, 448, 1, 576])]; tensor var_8785_end_mask_0 = const()[name = tensor("op_8785_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8785_cast_fp16 = slice_by_index(begin = var_8785_begin_0, end = var_8785_end_0, end_mask = var_8785_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8785_cast_fp16")]; tensor var_8789_begin_0 = const()[name = tensor("op_8789_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_8789_end_0 = const()[name = tensor("op_8789_end_0"), val = tensor([2, 512, 1, 576])]; tensor var_8789_end_mask_0 = const()[name = tensor("op_8789_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8789_cast_fp16 = slice_by_index(begin = var_8789_begin_0, end = var_8789_end_0, end_mask = var_8789_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8789_cast_fp16")]; tensor var_8793_begin_0 = const()[name = tensor("op_8793_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_8793_end_0 = const()[name = tensor("op_8793_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_8793_end_mask_0 = const()[name = tensor("op_8793_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8793_cast_fp16 = slice_by_index(begin = var_8793_begin_0, end = var_8793_end_0, end_mask = var_8793_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8793_cast_fp16")]; tensor var_8797_begin_0 = const()[name = tensor("op_8797_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_8797_end_0 = const()[name = tensor("op_8797_end_0"), val = tensor([2, 640, 1, 576])]; tensor var_8797_end_mask_0 = const()[name = tensor("op_8797_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8797_cast_fp16 = slice_by_index(begin = var_8797_begin_0, end = var_8797_end_0, end_mask = var_8797_end_mask_0, x = q_41_cast_fp16)[name = tensor("op_8797_cast_fp16")]; tensor k_83_perm_0 = const()[name = tensor("k_83_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_8804_begin_0 = const()[name = tensor("op_8804_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8804_end_0 = const()[name = tensor("op_8804_end_0"), val = tensor([2, 576, 1, 64])]; tensor var_8804_end_mask_0 = const()[name = tensor("op_8804_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_11 = transpose(perm = k_83_perm_0, x = k_81_cast_fp16)[name = tensor("transpose_11")]; tensor var_8804_cast_fp16 = slice_by_index(begin = var_8804_begin_0, end = var_8804_end_0, end_mask = var_8804_end_mask_0, x = transpose_11)[name = tensor("op_8804_cast_fp16")]; tensor var_8808_begin_0 = const()[name = tensor("op_8808_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_8808_end_0 = const()[name = tensor("op_8808_end_0"), val = tensor([2, 576, 1, 128])]; tensor var_8808_end_mask_0 = const()[name = tensor("op_8808_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8808_cast_fp16 = slice_by_index(begin = var_8808_begin_0, end = var_8808_end_0, end_mask = var_8808_end_mask_0, x = transpose_11)[name = tensor("op_8808_cast_fp16")]; tensor var_8812_begin_0 = const()[name = tensor("op_8812_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_8812_end_0 = const()[name = tensor("op_8812_end_0"), val = tensor([2, 576, 1, 192])]; tensor var_8812_end_mask_0 = const()[name = tensor("op_8812_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8812_cast_fp16 = slice_by_index(begin = var_8812_begin_0, end = var_8812_end_0, end_mask = var_8812_end_mask_0, x = transpose_11)[name = tensor("op_8812_cast_fp16")]; tensor var_8816_begin_0 = const()[name = tensor("op_8816_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_8816_end_0 = const()[name = tensor("op_8816_end_0"), val = tensor([2, 576, 1, 256])]; tensor var_8816_end_mask_0 = const()[name = tensor("op_8816_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8816_cast_fp16 = slice_by_index(begin = var_8816_begin_0, end = var_8816_end_0, end_mask = var_8816_end_mask_0, x = transpose_11)[name = tensor("op_8816_cast_fp16")]; tensor var_8820_begin_0 = const()[name = tensor("op_8820_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_8820_end_0 = const()[name = tensor("op_8820_end_0"), val = tensor([2, 576, 1, 320])]; tensor var_8820_end_mask_0 = const()[name = tensor("op_8820_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8820_cast_fp16 = slice_by_index(begin = var_8820_begin_0, end = var_8820_end_0, end_mask = var_8820_end_mask_0, x = transpose_11)[name = tensor("op_8820_cast_fp16")]; tensor var_8824_begin_0 = const()[name = tensor("op_8824_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_8824_end_0 = const()[name = tensor("op_8824_end_0"), val = tensor([2, 576, 1, 384])]; tensor var_8824_end_mask_0 = const()[name = tensor("op_8824_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8824_cast_fp16 = slice_by_index(begin = var_8824_begin_0, end = var_8824_end_0, end_mask = var_8824_end_mask_0, x = transpose_11)[name = tensor("op_8824_cast_fp16")]; tensor var_8828_begin_0 = const()[name = tensor("op_8828_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_8828_end_0 = const()[name = tensor("op_8828_end_0"), val = tensor([2, 576, 1, 448])]; tensor var_8828_end_mask_0 = const()[name = tensor("op_8828_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8828_cast_fp16 = slice_by_index(begin = var_8828_begin_0, end = var_8828_end_0, end_mask = var_8828_end_mask_0, x = transpose_11)[name = tensor("op_8828_cast_fp16")]; tensor var_8832_begin_0 = const()[name = tensor("op_8832_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_8832_end_0 = const()[name = tensor("op_8832_end_0"), val = tensor([2, 576, 1, 512])]; tensor var_8832_end_mask_0 = const()[name = tensor("op_8832_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8832_cast_fp16 = slice_by_index(begin = var_8832_begin_0, end = var_8832_end_0, end_mask = var_8832_end_mask_0, x = transpose_11)[name = tensor("op_8832_cast_fp16")]; tensor var_8836_begin_0 = const()[name = tensor("op_8836_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8836_end_0 = const()[name = tensor("op_8836_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_8836_end_mask_0 = const()[name = tensor("op_8836_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8836_cast_fp16 = slice_by_index(begin = var_8836_begin_0, end = var_8836_end_0, end_mask = var_8836_end_mask_0, x = transpose_11)[name = tensor("op_8836_cast_fp16")]; tensor var_8840_begin_0 = const()[name = tensor("op_8840_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_8840_end_0 = const()[name = tensor("op_8840_end_0"), val = tensor([2, 576, 1, 640])]; tensor var_8840_end_mask_0 = const()[name = tensor("op_8840_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8840_cast_fp16 = slice_by_index(begin = var_8840_begin_0, end = var_8840_end_0, end_mask = var_8840_end_mask_0, x = transpose_11)[name = tensor("op_8840_cast_fp16")]; tensor var_8842_begin_0 = const()[name = tensor("op_8842_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8842_end_0 = const()[name = tensor("op_8842_end_0"), val = tensor([2, 64, 1, 576])]; tensor var_8842_end_mask_0 = const()[name = tensor("op_8842_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8842_cast_fp16 = slice_by_index(begin = var_8842_begin_0, end = var_8842_end_0, end_mask = var_8842_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8842_cast_fp16")]; tensor var_8846_begin_0 = const()[name = tensor("op_8846_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_8846_end_0 = const()[name = tensor("op_8846_end_0"), val = tensor([2, 128, 1, 576])]; tensor var_8846_end_mask_0 = const()[name = tensor("op_8846_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8846_cast_fp16 = slice_by_index(begin = var_8846_begin_0, end = var_8846_end_0, end_mask = var_8846_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8846_cast_fp16")]; tensor var_8850_begin_0 = const()[name = tensor("op_8850_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_8850_end_0 = const()[name = tensor("op_8850_end_0"), val = tensor([2, 192, 1, 576])]; tensor var_8850_end_mask_0 = const()[name = tensor("op_8850_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8850_cast_fp16 = slice_by_index(begin = var_8850_begin_0, end = var_8850_end_0, end_mask = var_8850_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8850_cast_fp16")]; tensor var_8854_begin_0 = const()[name = tensor("op_8854_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_8854_end_0 = const()[name = tensor("op_8854_end_0"), val = tensor([2, 256, 1, 576])]; tensor var_8854_end_mask_0 = const()[name = tensor("op_8854_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8854_cast_fp16 = slice_by_index(begin = var_8854_begin_0, end = var_8854_end_0, end_mask = var_8854_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8854_cast_fp16")]; tensor var_8858_begin_0 = const()[name = tensor("op_8858_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_8858_end_0 = const()[name = tensor("op_8858_end_0"), val = tensor([2, 320, 1, 576])]; tensor var_8858_end_mask_0 = const()[name = tensor("op_8858_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8858_cast_fp16 = slice_by_index(begin = var_8858_begin_0, end = var_8858_end_0, end_mask = var_8858_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8858_cast_fp16")]; tensor var_8862_begin_0 = const()[name = tensor("op_8862_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8862_end_0 = const()[name = tensor("op_8862_end_0"), val = tensor([2, 384, 1, 576])]; tensor var_8862_end_mask_0 = const()[name = tensor("op_8862_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8862_cast_fp16 = slice_by_index(begin = var_8862_begin_0, end = var_8862_end_0, end_mask = var_8862_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8862_cast_fp16")]; tensor var_8866_begin_0 = const()[name = tensor("op_8866_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_8866_end_0 = const()[name = tensor("op_8866_end_0"), val = tensor([2, 448, 1, 576])]; tensor var_8866_end_mask_0 = const()[name = tensor("op_8866_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8866_cast_fp16 = slice_by_index(begin = var_8866_begin_0, end = var_8866_end_0, end_mask = var_8866_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8866_cast_fp16")]; tensor var_8870_begin_0 = const()[name = tensor("op_8870_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_8870_end_0 = const()[name = tensor("op_8870_end_0"), val = tensor([2, 512, 1, 576])]; tensor var_8870_end_mask_0 = const()[name = tensor("op_8870_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8870_cast_fp16 = slice_by_index(begin = var_8870_begin_0, end = var_8870_end_0, end_mask = var_8870_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8870_cast_fp16")]; tensor var_8874_begin_0 = const()[name = tensor("op_8874_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_8874_end_0 = const()[name = tensor("op_8874_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_8874_end_mask_0 = const()[name = tensor("op_8874_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8874_cast_fp16 = slice_by_index(begin = var_8874_begin_0, end = var_8874_end_0, end_mask = var_8874_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8874_cast_fp16")]; tensor var_8878_begin_0 = const()[name = tensor("op_8878_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_8878_end_0 = const()[name = tensor("op_8878_end_0"), val = tensor([2, 640, 1, 576])]; tensor var_8878_end_mask_0 = const()[name = tensor("op_8878_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8878_cast_fp16 = slice_by_index(begin = var_8878_begin_0, end = var_8878_end_0, end_mask = var_8878_end_mask_0, x = v_41_cast_fp16)[name = tensor("op_8878_cast_fp16")]; tensor var_8882_equation_0 = const()[name = tensor("op_8882_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8882_cast_fp16 = einsum(equation = var_8882_equation_0, values = (var_8804_cast_fp16, var_8761_cast_fp16))[name = tensor("op_8882_cast_fp16")]; tensor var_8883_to_fp16 = const()[name = tensor("op_8883_to_fp16"), val = tensor(0x1p-3)]; tensor aw_601_cast_fp16 = mul(x = var_8882_cast_fp16, y = var_8883_to_fp16)[name = tensor("aw_601_cast_fp16")]; tensor var_8886_equation_0 = const()[name = tensor("op_8886_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8886_cast_fp16 = einsum(equation = var_8886_equation_0, values = (var_8808_cast_fp16, var_8765_cast_fp16))[name = tensor("op_8886_cast_fp16")]; tensor var_8887_to_fp16 = const()[name = tensor("op_8887_to_fp16"), val = tensor(0x1p-3)]; tensor aw_603_cast_fp16 = mul(x = var_8886_cast_fp16, y = var_8887_to_fp16)[name = tensor("aw_603_cast_fp16")]; tensor var_8890_equation_0 = const()[name = tensor("op_8890_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8890_cast_fp16 = einsum(equation = var_8890_equation_0, values = (var_8812_cast_fp16, var_8769_cast_fp16))[name = tensor("op_8890_cast_fp16")]; tensor var_8891_to_fp16 = const()[name = tensor("op_8891_to_fp16"), val = tensor(0x1p-3)]; tensor aw_605_cast_fp16 = mul(x = var_8890_cast_fp16, y = var_8891_to_fp16)[name = tensor("aw_605_cast_fp16")]; tensor var_8894_equation_0 = const()[name = tensor("op_8894_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8894_cast_fp16 = einsum(equation = var_8894_equation_0, values = (var_8816_cast_fp16, var_8773_cast_fp16))[name = tensor("op_8894_cast_fp16")]; tensor var_8895_to_fp16 = const()[name = tensor("op_8895_to_fp16"), val = tensor(0x1p-3)]; tensor aw_607_cast_fp16 = mul(x = var_8894_cast_fp16, y = var_8895_to_fp16)[name = tensor("aw_607_cast_fp16")]; tensor var_8898_equation_0 = const()[name = tensor("op_8898_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8898_cast_fp16 = einsum(equation = var_8898_equation_0, values = (var_8820_cast_fp16, var_8777_cast_fp16))[name = tensor("op_8898_cast_fp16")]; tensor var_8899_to_fp16 = const()[name = tensor("op_8899_to_fp16"), val = tensor(0x1p-3)]; tensor aw_609_cast_fp16 = mul(x = var_8898_cast_fp16, y = var_8899_to_fp16)[name = tensor("aw_609_cast_fp16")]; tensor var_8902_equation_0 = const()[name = tensor("op_8902_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8902_cast_fp16 = einsum(equation = var_8902_equation_0, values = (var_8824_cast_fp16, var_8781_cast_fp16))[name = tensor("op_8902_cast_fp16")]; tensor var_8903_to_fp16 = const()[name = tensor("op_8903_to_fp16"), val = tensor(0x1p-3)]; tensor aw_611_cast_fp16 = mul(x = var_8902_cast_fp16, y = var_8903_to_fp16)[name = tensor("aw_611_cast_fp16")]; tensor var_8906_equation_0 = const()[name = tensor("op_8906_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8906_cast_fp16 = einsum(equation = var_8906_equation_0, values = (var_8828_cast_fp16, var_8785_cast_fp16))[name = tensor("op_8906_cast_fp16")]; tensor var_8907_to_fp16 = const()[name = tensor("op_8907_to_fp16"), val = tensor(0x1p-3)]; tensor aw_613_cast_fp16 = mul(x = var_8906_cast_fp16, y = var_8907_to_fp16)[name = tensor("aw_613_cast_fp16")]; tensor var_8910_equation_0 = const()[name = tensor("op_8910_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8910_cast_fp16 = einsum(equation = var_8910_equation_0, values = (var_8832_cast_fp16, var_8789_cast_fp16))[name = tensor("op_8910_cast_fp16")]; tensor var_8911_to_fp16 = const()[name = tensor("op_8911_to_fp16"), val = tensor(0x1p-3)]; tensor aw_615_cast_fp16 = mul(x = var_8910_cast_fp16, y = var_8911_to_fp16)[name = tensor("aw_615_cast_fp16")]; tensor var_8914_equation_0 = const()[name = tensor("op_8914_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8914_cast_fp16 = einsum(equation = var_8914_equation_0, values = (var_8836_cast_fp16, var_8793_cast_fp16))[name = tensor("op_8914_cast_fp16")]; tensor var_8915_to_fp16 = const()[name = tensor("op_8915_to_fp16"), val = tensor(0x1p-3)]; tensor aw_617_cast_fp16 = mul(x = var_8914_cast_fp16, y = var_8915_to_fp16)[name = tensor("aw_617_cast_fp16")]; tensor var_8918_equation_0 = const()[name = tensor("op_8918_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8918_cast_fp16 = einsum(equation = var_8918_equation_0, values = (var_8840_cast_fp16, var_8797_cast_fp16))[name = tensor("op_8918_cast_fp16")]; tensor var_8919_to_fp16 = const()[name = tensor("op_8919_to_fp16"), val = tensor(0x1p-3)]; tensor aw_619_cast_fp16 = mul(x = var_8918_cast_fp16, y = var_8919_to_fp16)[name = tensor("aw_619_cast_fp16")]; tensor var_8921_cast_fp16 = softmax(axis = var_8627, x = aw_601_cast_fp16)[name = tensor("op_8921_cast_fp16")]; tensor var_8922_cast_fp16 = softmax(axis = var_8627, x = aw_603_cast_fp16)[name = tensor("op_8922_cast_fp16")]; tensor var_8923_cast_fp16 = softmax(axis = var_8627, x = aw_605_cast_fp16)[name = tensor("op_8923_cast_fp16")]; tensor var_8924_cast_fp16 = softmax(axis = var_8627, x = aw_607_cast_fp16)[name = tensor("op_8924_cast_fp16")]; tensor var_8925_cast_fp16 = softmax(axis = var_8627, x = aw_609_cast_fp16)[name = tensor("op_8925_cast_fp16")]; tensor var_8926_cast_fp16 = softmax(axis = var_8627, x = aw_611_cast_fp16)[name = tensor("op_8926_cast_fp16")]; tensor var_8927_cast_fp16 = softmax(axis = var_8627, x = aw_613_cast_fp16)[name = tensor("op_8927_cast_fp16")]; tensor var_8928_cast_fp16 = softmax(axis = var_8627, x = aw_615_cast_fp16)[name = tensor("op_8928_cast_fp16")]; tensor var_8929_cast_fp16 = softmax(axis = var_8627, x = aw_617_cast_fp16)[name = tensor("op_8929_cast_fp16")]; tensor var_8930_cast_fp16 = softmax(axis = var_8627, x = aw_619_cast_fp16)[name = tensor("op_8930_cast_fp16")]; tensor var_8932_equation_0 = const()[name = tensor("op_8932_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8932_cast_fp16 = einsum(equation = var_8932_equation_0, values = (var_8842_cast_fp16, var_8921_cast_fp16))[name = tensor("op_8932_cast_fp16")]; tensor var_8934_equation_0 = const()[name = tensor("op_8934_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8934_cast_fp16 = einsum(equation = var_8934_equation_0, values = (var_8846_cast_fp16, var_8922_cast_fp16))[name = tensor("op_8934_cast_fp16")]; tensor var_8936_equation_0 = const()[name = tensor("op_8936_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8936_cast_fp16 = einsum(equation = var_8936_equation_0, values = (var_8850_cast_fp16, var_8923_cast_fp16))[name = tensor("op_8936_cast_fp16")]; tensor var_8938_equation_0 = const()[name = tensor("op_8938_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8938_cast_fp16 = einsum(equation = var_8938_equation_0, values = (var_8854_cast_fp16, var_8924_cast_fp16))[name = tensor("op_8938_cast_fp16")]; tensor var_8940_equation_0 = const()[name = tensor("op_8940_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8940_cast_fp16 = einsum(equation = var_8940_equation_0, values = (var_8858_cast_fp16, var_8925_cast_fp16))[name = tensor("op_8940_cast_fp16")]; tensor var_8942_equation_0 = const()[name = tensor("op_8942_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8942_cast_fp16 = einsum(equation = var_8942_equation_0, values = (var_8862_cast_fp16, var_8926_cast_fp16))[name = tensor("op_8942_cast_fp16")]; tensor var_8944_equation_0 = const()[name = tensor("op_8944_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8944_cast_fp16 = einsum(equation = var_8944_equation_0, values = (var_8866_cast_fp16, var_8927_cast_fp16))[name = tensor("op_8944_cast_fp16")]; tensor var_8946_equation_0 = const()[name = tensor("op_8946_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8946_cast_fp16 = einsum(equation = var_8946_equation_0, values = (var_8870_cast_fp16, var_8928_cast_fp16))[name = tensor("op_8946_cast_fp16")]; tensor var_8948_equation_0 = const()[name = tensor("op_8948_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8948_cast_fp16 = einsum(equation = var_8948_equation_0, values = (var_8874_cast_fp16, var_8929_cast_fp16))[name = tensor("op_8948_cast_fp16")]; tensor var_8950_equation_0 = const()[name = tensor("op_8950_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8950_cast_fp16 = einsum(equation = var_8950_equation_0, values = (var_8878_cast_fp16, var_8930_cast_fp16))[name = tensor("op_8950_cast_fp16")]; tensor input_383_interleave_0 = const()[name = tensor("input_383_interleave_0"), val = tensor(false)]; tensor input_383_cast_fp16 = concat(axis = var_8627, interleave = input_383_interleave_0, values = (var_8932_cast_fp16, var_8934_cast_fp16, var_8936_cast_fp16, var_8938_cast_fp16, var_8940_cast_fp16, var_8942_cast_fp16, var_8944_cast_fp16, var_8946_cast_fp16, var_8948_cast_fp16, var_8950_cast_fp16))[name = tensor("input_383_cast_fp16")]; tensor var_8956 = const()[name = tensor("op_8956"), val = tensor([1, 1])]; tensor var_8958 = const()[name = tensor("op_8958"), val = tensor([1, 1])]; tensor var_8960_pad_type_0 = const()[name = tensor("op_8960_pad_type_0"), val = tensor("custom")]; tensor var_8960_pad_0 = const()[name = tensor("op_8960_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(691896576)))]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(692715840)))]; tensor var_8960_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_8958, groups = var_8627, pad = var_8960_pad_0, pad_type = var_8960_pad_type_0, strides = var_8956, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_383_cast_fp16)[name = tensor("op_8960_cast_fp16")]; tensor inputs_63_cast_fp16 = add(x = var_8960_cast_fp16, y = inputs_61_cast_fp16)[name = tensor("inputs_63_cast_fp16")]; tensor var_8964 = const()[name = tensor("op_8964"), val = tensor([1])]; tensor channels_mean_63_cast_fp16 = reduce_mean(axes = var_8964, keep_dims = var_8622, x = inputs_63_cast_fp16)[name = tensor("channels_mean_63_cast_fp16")]; tensor zero_mean_63_cast_fp16 = sub(x = inputs_63_cast_fp16, y = channels_mean_63_cast_fp16)[name = tensor("zero_mean_63_cast_fp16")]; tensor zero_mean_sq_63_cast_fp16 = mul(x = zero_mean_63_cast_fp16, y = zero_mean_63_cast_fp16)[name = tensor("zero_mean_sq_63_cast_fp16")]; tensor var_8968 = const()[name = tensor("op_8968"), val = tensor([1])]; tensor var_8969_cast_fp16 = reduce_mean(axes = var_8968, keep_dims = var_8622, x = zero_mean_sq_63_cast_fp16)[name = tensor("op_8969_cast_fp16")]; tensor var_8970_to_fp16 = const()[name = tensor("op_8970_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_8971_cast_fp16 = add(x = var_8969_cast_fp16, y = var_8970_to_fp16)[name = tensor("op_8971_cast_fp16")]; tensor denom_63_epsilon_0_to_fp16 = const()[name = tensor("denom_63_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_63_cast_fp16 = rsqrt(epsilon = denom_63_epsilon_0_to_fp16, x = var_8971_cast_fp16)[name = tensor("denom_63_cast_fp16")]; tensor out_63_cast_fp16 = mul(x = zero_mean_63_cast_fp16, y = denom_63_cast_fp16)[name = tensor("out_63_cast_fp16")]; tensor var_8975_to_fp16 = const()[name = tensor("op_8975_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(692717184)))]; tensor var_8976_cast_fp16 = add(x = out_63_cast_fp16, y = var_8975_to_fp16)[name = tensor("op_8976_cast_fp16")]; tensor var_8978_to_fp16 = const()[name = tensor("op_8978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(692718528)))]; tensor hidden_states_229_cast_fp16 = mul(x = var_8976_cast_fp16, y = var_8978_to_fp16)[name = tensor("hidden_states_229_cast_fp16")]; tensor var_8985 = const()[name = tensor("op_8985"), val = tensor([1, 1])]; tensor var_8987 = const()[name = tensor("op_8987"), val = tensor([1, 1])]; tensor q_43_pad_type_0 = const()[name = tensor("q_43_pad_type_0"), val = tensor("custom")]; tensor q_43_pad_0 = const()[name = tensor("q_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(692719872)))]; tensor q_43_cast_fp16 = conv(dilations = var_8987, groups = var_8627, pad = q_43_pad_0, pad_type = q_43_pad_type_0, strides = var_8985, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_229_cast_fp16)[name = tensor("q_43_cast_fp16")]; tensor var_8991 = const()[name = tensor("op_8991"), val = tensor([1, 1])]; tensor var_8993 = const()[name = tensor("op_8993"), val = tensor([1, 1])]; tensor k_85_pad_type_0 = const()[name = tensor("k_85_pad_type_0"), val = tensor("custom")]; tensor k_85_pad_0 = const()[name = tensor("k_85_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693539136)))]; tensor k_85_cast_fp16 = conv(dilations = var_8993, groups = var_8627, pad = k_85_pad_0, pad_type = k_85_pad_type_0, strides = var_8991, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_85_cast_fp16")]; tensor var_8997 = const()[name = tensor("op_8997"), val = tensor([1, 1])]; tensor var_8999 = const()[name = tensor("op_8999"), val = tensor([1, 1])]; tensor v_43_pad_type_0 = const()[name = tensor("v_43_pad_type_0"), val = tensor("custom")]; tensor v_43_pad_0 = const()[name = tensor("v_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(694849920)))]; tensor v_43_cast_fp16 = conv(dilations = var_8999, groups = var_8627, pad = v_43_pad_0, pad_type = v_43_pad_type_0, strides = var_8997, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_43_cast_fp16")]; tensor var_9003_begin_0 = const()[name = tensor("op_9003_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9003_end_0 = const()[name = tensor("op_9003_end_0"), val = tensor([2, 64, 1, 576])]; tensor var_9003_end_mask_0 = const()[name = tensor("op_9003_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9003_cast_fp16 = slice_by_index(begin = var_9003_begin_0, end = var_9003_end_0, end_mask = var_9003_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_9003_cast_fp16")]; tensor var_9007_begin_0 = const()[name = tensor("op_9007_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_9007_end_0 = const()[name = tensor("op_9007_end_0"), val = tensor([2, 128, 1, 576])]; tensor var_9007_end_mask_0 = const()[name = tensor("op_9007_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9007_cast_fp16 = slice_by_index(begin = var_9007_begin_0, end = var_9007_end_0, end_mask = var_9007_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_9007_cast_fp16")]; tensor var_9011_begin_0 = const()[name = tensor("op_9011_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_9011_end_0 = const()[name = tensor("op_9011_end_0"), val = tensor([2, 192, 1, 576])]; tensor var_9011_end_mask_0 = const()[name = tensor("op_9011_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9011_cast_fp16 = slice_by_index(begin = var_9011_begin_0, end = var_9011_end_0, end_mask = var_9011_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_9011_cast_fp16")]; tensor var_9015_begin_0 = const()[name = tensor("op_9015_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_9015_end_0 = const()[name = tensor("op_9015_end_0"), val = tensor([2, 256, 1, 576])]; tensor var_9015_end_mask_0 = const()[name = tensor("op_9015_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9015_cast_fp16 = slice_by_index(begin = var_9015_begin_0, end = var_9015_end_0, end_mask = var_9015_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_9015_cast_fp16")]; tensor var_9019_begin_0 = const()[name = tensor("op_9019_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_9019_end_0 = const()[name = tensor("op_9019_end_0"), val = tensor([2, 320, 1, 576])]; tensor var_9019_end_mask_0 = const()[name = tensor("op_9019_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9019_cast_fp16 = slice_by_index(begin = var_9019_begin_0, end = var_9019_end_0, end_mask = var_9019_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_9019_cast_fp16")]; tensor var_9023_begin_0 = const()[name = tensor("op_9023_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9023_end_0 = const()[name = tensor("op_9023_end_0"), val = tensor([2, 384, 1, 576])]; tensor var_9023_end_mask_0 = const()[name = tensor("op_9023_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9023_cast_fp16 = slice_by_index(begin = var_9023_begin_0, end = var_9023_end_0, end_mask = var_9023_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_9023_cast_fp16")]; tensor var_9027_begin_0 = const()[name = tensor("op_9027_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_9027_end_0 = const()[name = tensor("op_9027_end_0"), val = tensor([2, 448, 1, 576])]; tensor var_9027_end_mask_0 = const()[name = tensor("op_9027_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9027_cast_fp16 = slice_by_index(begin = var_9027_begin_0, end = var_9027_end_0, end_mask = var_9027_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_9027_cast_fp16")]; tensor var_9031_begin_0 = const()[name = tensor("op_9031_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_9031_end_0 = const()[name = tensor("op_9031_end_0"), val = tensor([2, 512, 1, 576])]; tensor var_9031_end_mask_0 = const()[name = tensor("op_9031_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9031_cast_fp16 = slice_by_index(begin = var_9031_begin_0, end = var_9031_end_0, end_mask = var_9031_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_9031_cast_fp16")]; tensor var_9035_begin_0 = const()[name = tensor("op_9035_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_9035_end_0 = const()[name = tensor("op_9035_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_9035_end_mask_0 = const()[name = tensor("op_9035_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9035_cast_fp16 = slice_by_index(begin = var_9035_begin_0, end = var_9035_end_0, end_mask = var_9035_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_9035_cast_fp16")]; tensor var_9039_begin_0 = const()[name = tensor("op_9039_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_9039_end_0 = const()[name = tensor("op_9039_end_0"), val = tensor([2, 640, 1, 576])]; tensor var_9039_end_mask_0 = const()[name = tensor("op_9039_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9039_cast_fp16 = slice_by_index(begin = var_9039_begin_0, end = var_9039_end_0, end_mask = var_9039_end_mask_0, x = q_43_cast_fp16)[name = tensor("op_9039_cast_fp16")]; tensor k_87_perm_0 = const()[name = tensor("k_87_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_9046_begin_0 = const()[name = tensor("op_9046_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9046_end_0 = const()[name = tensor("op_9046_end_0"), val = tensor([2, 77, 1, 64])]; tensor var_9046_end_mask_0 = const()[name = tensor("op_9046_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_10 = transpose(perm = k_87_perm_0, x = k_85_cast_fp16)[name = tensor("transpose_10")]; tensor var_9046_cast_fp16 = slice_by_index(begin = var_9046_begin_0, end = var_9046_end_0, end_mask = var_9046_end_mask_0, x = transpose_10)[name = tensor("op_9046_cast_fp16")]; tensor var_9050_begin_0 = const()[name = tensor("op_9050_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_9050_end_0 = const()[name = tensor("op_9050_end_0"), val = tensor([2, 77, 1, 128])]; tensor var_9050_end_mask_0 = const()[name = tensor("op_9050_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9050_cast_fp16 = slice_by_index(begin = var_9050_begin_0, end = var_9050_end_0, end_mask = var_9050_end_mask_0, x = transpose_10)[name = tensor("op_9050_cast_fp16")]; tensor var_9054_begin_0 = const()[name = tensor("op_9054_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_9054_end_0 = const()[name = tensor("op_9054_end_0"), val = tensor([2, 77, 1, 192])]; tensor var_9054_end_mask_0 = const()[name = tensor("op_9054_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9054_cast_fp16 = slice_by_index(begin = var_9054_begin_0, end = var_9054_end_0, end_mask = var_9054_end_mask_0, x = transpose_10)[name = tensor("op_9054_cast_fp16")]; tensor var_9058_begin_0 = const()[name = tensor("op_9058_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_9058_end_0 = const()[name = tensor("op_9058_end_0"), val = tensor([2, 77, 1, 256])]; tensor var_9058_end_mask_0 = const()[name = tensor("op_9058_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9058_cast_fp16 = slice_by_index(begin = var_9058_begin_0, end = var_9058_end_0, end_mask = var_9058_end_mask_0, x = transpose_10)[name = tensor("op_9058_cast_fp16")]; tensor var_9062_begin_0 = const()[name = tensor("op_9062_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_9062_end_0 = const()[name = tensor("op_9062_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_9062_end_mask_0 = const()[name = tensor("op_9062_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9062_cast_fp16 = slice_by_index(begin = var_9062_begin_0, end = var_9062_end_0, end_mask = var_9062_end_mask_0, x = transpose_10)[name = tensor("op_9062_cast_fp16")]; tensor var_9066_begin_0 = const()[name = tensor("op_9066_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_9066_end_0 = const()[name = tensor("op_9066_end_0"), val = tensor([2, 77, 1, 384])]; tensor var_9066_end_mask_0 = const()[name = tensor("op_9066_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9066_cast_fp16 = slice_by_index(begin = var_9066_begin_0, end = var_9066_end_0, end_mask = var_9066_end_mask_0, x = transpose_10)[name = tensor("op_9066_cast_fp16")]; tensor var_9070_begin_0 = const()[name = tensor("op_9070_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_9070_end_0 = const()[name = tensor("op_9070_end_0"), val = tensor([2, 77, 1, 448])]; tensor var_9070_end_mask_0 = const()[name = tensor("op_9070_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9070_cast_fp16 = slice_by_index(begin = var_9070_begin_0, end = var_9070_end_0, end_mask = var_9070_end_mask_0, x = transpose_10)[name = tensor("op_9070_cast_fp16")]; tensor var_9074_begin_0 = const()[name = tensor("op_9074_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_9074_end_0 = const()[name = tensor("op_9074_end_0"), val = tensor([2, 77, 1, 512])]; tensor var_9074_end_mask_0 = const()[name = tensor("op_9074_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9074_cast_fp16 = slice_by_index(begin = var_9074_begin_0, end = var_9074_end_0, end_mask = var_9074_end_mask_0, x = transpose_10)[name = tensor("op_9074_cast_fp16")]; tensor var_9078_begin_0 = const()[name = tensor("op_9078_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9078_end_0 = const()[name = tensor("op_9078_end_0"), val = tensor([2, 77, 1, 576])]; tensor var_9078_end_mask_0 = const()[name = tensor("op_9078_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9078_cast_fp16 = slice_by_index(begin = var_9078_begin_0, end = var_9078_end_0, end_mask = var_9078_end_mask_0, x = transpose_10)[name = tensor("op_9078_cast_fp16")]; tensor var_9082_begin_0 = const()[name = tensor("op_9082_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_9082_end_0 = const()[name = tensor("op_9082_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_9082_end_mask_0 = const()[name = tensor("op_9082_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9082_cast_fp16 = slice_by_index(begin = var_9082_begin_0, end = var_9082_end_0, end_mask = var_9082_end_mask_0, x = transpose_10)[name = tensor("op_9082_cast_fp16")]; tensor var_9084_begin_0 = const()[name = tensor("op_9084_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9084_end_0 = const()[name = tensor("op_9084_end_0"), val = tensor([2, 64, 1, 77])]; tensor var_9084_end_mask_0 = const()[name = tensor("op_9084_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9084_cast_fp16 = slice_by_index(begin = var_9084_begin_0, end = var_9084_end_0, end_mask = var_9084_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_9084_cast_fp16")]; tensor var_9088_begin_0 = const()[name = tensor("op_9088_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_9088_end_0 = const()[name = tensor("op_9088_end_0"), val = tensor([2, 128, 1, 77])]; tensor var_9088_end_mask_0 = const()[name = tensor("op_9088_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9088_cast_fp16 = slice_by_index(begin = var_9088_begin_0, end = var_9088_end_0, end_mask = var_9088_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_9088_cast_fp16")]; tensor var_9092_begin_0 = const()[name = tensor("op_9092_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_9092_end_0 = const()[name = tensor("op_9092_end_0"), val = tensor([2, 192, 1, 77])]; tensor var_9092_end_mask_0 = const()[name = tensor("op_9092_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9092_cast_fp16 = slice_by_index(begin = var_9092_begin_0, end = var_9092_end_0, end_mask = var_9092_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_9092_cast_fp16")]; tensor var_9096_begin_0 = const()[name = tensor("op_9096_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_9096_end_0 = const()[name = tensor("op_9096_end_0"), val = tensor([2, 256, 1, 77])]; tensor var_9096_end_mask_0 = const()[name = tensor("op_9096_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9096_cast_fp16 = slice_by_index(begin = var_9096_begin_0, end = var_9096_end_0, end_mask = var_9096_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_9096_cast_fp16")]; tensor var_9100_begin_0 = const()[name = tensor("op_9100_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_9100_end_0 = const()[name = tensor("op_9100_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_9100_end_mask_0 = const()[name = tensor("op_9100_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9100_cast_fp16 = slice_by_index(begin = var_9100_begin_0, end = var_9100_end_0, end_mask = var_9100_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_9100_cast_fp16")]; tensor var_9104_begin_0 = const()[name = tensor("op_9104_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9104_end_0 = const()[name = tensor("op_9104_end_0"), val = tensor([2, 384, 1, 77])]; tensor var_9104_end_mask_0 = const()[name = tensor("op_9104_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9104_cast_fp16 = slice_by_index(begin = var_9104_begin_0, end = var_9104_end_0, end_mask = var_9104_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_9104_cast_fp16")]; tensor var_9108_begin_0 = const()[name = tensor("op_9108_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_9108_end_0 = const()[name = tensor("op_9108_end_0"), val = tensor([2, 448, 1, 77])]; tensor var_9108_end_mask_0 = const()[name = tensor("op_9108_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9108_cast_fp16 = slice_by_index(begin = var_9108_begin_0, end = var_9108_end_0, end_mask = var_9108_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_9108_cast_fp16")]; tensor var_9112_begin_0 = const()[name = tensor("op_9112_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_9112_end_0 = const()[name = tensor("op_9112_end_0"), val = tensor([2, 512, 1, 77])]; tensor var_9112_end_mask_0 = const()[name = tensor("op_9112_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9112_cast_fp16 = slice_by_index(begin = var_9112_begin_0, end = var_9112_end_0, end_mask = var_9112_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_9112_cast_fp16")]; tensor var_9116_begin_0 = const()[name = tensor("op_9116_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_9116_end_0 = const()[name = tensor("op_9116_end_0"), val = tensor([2, 576, 1, 77])]; tensor var_9116_end_mask_0 = const()[name = tensor("op_9116_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9116_cast_fp16 = slice_by_index(begin = var_9116_begin_0, end = var_9116_end_0, end_mask = var_9116_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_9116_cast_fp16")]; tensor var_9120_begin_0 = const()[name = tensor("op_9120_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_9120_end_0 = const()[name = tensor("op_9120_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_9120_end_mask_0 = const()[name = tensor("op_9120_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9120_cast_fp16 = slice_by_index(begin = var_9120_begin_0, end = var_9120_end_0, end_mask = var_9120_end_mask_0, x = v_43_cast_fp16)[name = tensor("op_9120_cast_fp16")]; tensor var_9124_equation_0 = const()[name = tensor("op_9124_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9124_cast_fp16 = einsum(equation = var_9124_equation_0, values = (var_9046_cast_fp16, var_9003_cast_fp16))[name = tensor("op_9124_cast_fp16")]; tensor var_9125_to_fp16 = const()[name = tensor("op_9125_to_fp16"), val = tensor(0x1p-3)]; tensor aw_621_cast_fp16 = mul(x = var_9124_cast_fp16, y = var_9125_to_fp16)[name = tensor("aw_621_cast_fp16")]; tensor var_9128_equation_0 = const()[name = tensor("op_9128_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9128_cast_fp16 = einsum(equation = var_9128_equation_0, values = (var_9050_cast_fp16, var_9007_cast_fp16))[name = tensor("op_9128_cast_fp16")]; tensor var_9129_to_fp16 = const()[name = tensor("op_9129_to_fp16"), val = tensor(0x1p-3)]; tensor aw_623_cast_fp16 = mul(x = var_9128_cast_fp16, y = var_9129_to_fp16)[name = tensor("aw_623_cast_fp16")]; tensor var_9132_equation_0 = const()[name = tensor("op_9132_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9132_cast_fp16 = einsum(equation = var_9132_equation_0, values = (var_9054_cast_fp16, var_9011_cast_fp16))[name = tensor("op_9132_cast_fp16")]; tensor var_9133_to_fp16 = const()[name = tensor("op_9133_to_fp16"), val = tensor(0x1p-3)]; tensor aw_625_cast_fp16 = mul(x = var_9132_cast_fp16, y = var_9133_to_fp16)[name = tensor("aw_625_cast_fp16")]; tensor var_9136_equation_0 = const()[name = tensor("op_9136_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9136_cast_fp16 = einsum(equation = var_9136_equation_0, values = (var_9058_cast_fp16, var_9015_cast_fp16))[name = tensor("op_9136_cast_fp16")]; tensor var_9137_to_fp16 = const()[name = tensor("op_9137_to_fp16"), val = tensor(0x1p-3)]; tensor aw_627_cast_fp16 = mul(x = var_9136_cast_fp16, y = var_9137_to_fp16)[name = tensor("aw_627_cast_fp16")]; tensor var_9140_equation_0 = const()[name = tensor("op_9140_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9140_cast_fp16 = einsum(equation = var_9140_equation_0, values = (var_9062_cast_fp16, var_9019_cast_fp16))[name = tensor("op_9140_cast_fp16")]; tensor var_9141_to_fp16 = const()[name = tensor("op_9141_to_fp16"), val = tensor(0x1p-3)]; tensor aw_629_cast_fp16 = mul(x = var_9140_cast_fp16, y = var_9141_to_fp16)[name = tensor("aw_629_cast_fp16")]; tensor var_9144_equation_0 = const()[name = tensor("op_9144_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9144_cast_fp16 = einsum(equation = var_9144_equation_0, values = (var_9066_cast_fp16, var_9023_cast_fp16))[name = tensor("op_9144_cast_fp16")]; tensor var_9145_to_fp16 = const()[name = tensor("op_9145_to_fp16"), val = tensor(0x1p-3)]; tensor aw_631_cast_fp16 = mul(x = var_9144_cast_fp16, y = var_9145_to_fp16)[name = tensor("aw_631_cast_fp16")]; tensor var_9148_equation_0 = const()[name = tensor("op_9148_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9148_cast_fp16 = einsum(equation = var_9148_equation_0, values = (var_9070_cast_fp16, var_9027_cast_fp16))[name = tensor("op_9148_cast_fp16")]; tensor var_9149_to_fp16 = const()[name = tensor("op_9149_to_fp16"), val = tensor(0x1p-3)]; tensor aw_633_cast_fp16 = mul(x = var_9148_cast_fp16, y = var_9149_to_fp16)[name = tensor("aw_633_cast_fp16")]; tensor var_9152_equation_0 = const()[name = tensor("op_9152_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9152_cast_fp16 = einsum(equation = var_9152_equation_0, values = (var_9074_cast_fp16, var_9031_cast_fp16))[name = tensor("op_9152_cast_fp16")]; tensor var_9153_to_fp16 = const()[name = tensor("op_9153_to_fp16"), val = tensor(0x1p-3)]; tensor aw_635_cast_fp16 = mul(x = var_9152_cast_fp16, y = var_9153_to_fp16)[name = tensor("aw_635_cast_fp16")]; tensor var_9156_equation_0 = const()[name = tensor("op_9156_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9156_cast_fp16 = einsum(equation = var_9156_equation_0, values = (var_9078_cast_fp16, var_9035_cast_fp16))[name = tensor("op_9156_cast_fp16")]; tensor var_9157_to_fp16 = const()[name = tensor("op_9157_to_fp16"), val = tensor(0x1p-3)]; tensor aw_637_cast_fp16 = mul(x = var_9156_cast_fp16, y = var_9157_to_fp16)[name = tensor("aw_637_cast_fp16")]; tensor var_9160_equation_0 = const()[name = tensor("op_9160_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9160_cast_fp16 = einsum(equation = var_9160_equation_0, values = (var_9082_cast_fp16, var_9039_cast_fp16))[name = tensor("op_9160_cast_fp16")]; tensor var_9161_to_fp16 = const()[name = tensor("op_9161_to_fp16"), val = tensor(0x1p-3)]; tensor aw_639_cast_fp16 = mul(x = var_9160_cast_fp16, y = var_9161_to_fp16)[name = tensor("aw_639_cast_fp16")]; tensor var_9163_cast_fp16 = softmax(axis = var_8627, x = aw_621_cast_fp16)[name = tensor("op_9163_cast_fp16")]; tensor var_9164_cast_fp16 = softmax(axis = var_8627, x = aw_623_cast_fp16)[name = tensor("op_9164_cast_fp16")]; tensor var_9165_cast_fp16 = softmax(axis = var_8627, x = aw_625_cast_fp16)[name = tensor("op_9165_cast_fp16")]; tensor var_9166_cast_fp16 = softmax(axis = var_8627, x = aw_627_cast_fp16)[name = tensor("op_9166_cast_fp16")]; tensor var_9167_cast_fp16 = softmax(axis = var_8627, x = aw_629_cast_fp16)[name = tensor("op_9167_cast_fp16")]; tensor var_9168_cast_fp16 = softmax(axis = var_8627, x = aw_631_cast_fp16)[name = tensor("op_9168_cast_fp16")]; tensor var_9169_cast_fp16 = softmax(axis = var_8627, x = aw_633_cast_fp16)[name = tensor("op_9169_cast_fp16")]; tensor var_9170_cast_fp16 = softmax(axis = var_8627, x = aw_635_cast_fp16)[name = tensor("op_9170_cast_fp16")]; tensor var_9171_cast_fp16 = softmax(axis = var_8627, x = aw_637_cast_fp16)[name = tensor("op_9171_cast_fp16")]; tensor var_9172_cast_fp16 = softmax(axis = var_8627, x = aw_639_cast_fp16)[name = tensor("op_9172_cast_fp16")]; tensor var_9174_equation_0 = const()[name = tensor("op_9174_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9174_cast_fp16 = einsum(equation = var_9174_equation_0, values = (var_9084_cast_fp16, var_9163_cast_fp16))[name = tensor("op_9174_cast_fp16")]; tensor var_9176_equation_0 = const()[name = tensor("op_9176_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9176_cast_fp16 = einsum(equation = var_9176_equation_0, values = (var_9088_cast_fp16, var_9164_cast_fp16))[name = tensor("op_9176_cast_fp16")]; tensor var_9178_equation_0 = const()[name = tensor("op_9178_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9178_cast_fp16 = einsum(equation = var_9178_equation_0, values = (var_9092_cast_fp16, var_9165_cast_fp16))[name = tensor("op_9178_cast_fp16")]; tensor var_9180_equation_0 = const()[name = tensor("op_9180_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9180_cast_fp16 = einsum(equation = var_9180_equation_0, values = (var_9096_cast_fp16, var_9166_cast_fp16))[name = tensor("op_9180_cast_fp16")]; tensor var_9182_equation_0 = const()[name = tensor("op_9182_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9182_cast_fp16 = einsum(equation = var_9182_equation_0, values = (var_9100_cast_fp16, var_9167_cast_fp16))[name = tensor("op_9182_cast_fp16")]; tensor var_9184_equation_0 = const()[name = tensor("op_9184_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9184_cast_fp16 = einsum(equation = var_9184_equation_0, values = (var_9104_cast_fp16, var_9168_cast_fp16))[name = tensor("op_9184_cast_fp16")]; tensor var_9186_equation_0 = const()[name = tensor("op_9186_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9186_cast_fp16 = einsum(equation = var_9186_equation_0, values = (var_9108_cast_fp16, var_9169_cast_fp16))[name = tensor("op_9186_cast_fp16")]; tensor var_9188_equation_0 = const()[name = tensor("op_9188_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9188_cast_fp16 = einsum(equation = var_9188_equation_0, values = (var_9112_cast_fp16, var_9170_cast_fp16))[name = tensor("op_9188_cast_fp16")]; tensor var_9190_equation_0 = const()[name = tensor("op_9190_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9190_cast_fp16 = einsum(equation = var_9190_equation_0, values = (var_9116_cast_fp16, var_9171_cast_fp16))[name = tensor("op_9190_cast_fp16")]; tensor var_9192_equation_0 = const()[name = tensor("op_9192_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9192_cast_fp16 = einsum(equation = var_9192_equation_0, values = (var_9120_cast_fp16, var_9172_cast_fp16))[name = tensor("op_9192_cast_fp16")]; tensor input_385_interleave_0 = const()[name = tensor("input_385_interleave_0"), val = tensor(false)]; tensor input_385_cast_fp16 = concat(axis = var_8627, interleave = input_385_interleave_0, values = (var_9174_cast_fp16, var_9176_cast_fp16, var_9178_cast_fp16, var_9180_cast_fp16, var_9182_cast_fp16, var_9184_cast_fp16, var_9186_cast_fp16, var_9188_cast_fp16, var_9190_cast_fp16, var_9192_cast_fp16))[name = tensor("input_385_cast_fp16")]; tensor var_9198 = const()[name = tensor("op_9198"), val = tensor([1, 1])]; tensor var_9200 = const()[name = tensor("op_9200"), val = tensor([1, 1])]; tensor var_9202_pad_type_0 = const()[name = tensor("op_9202_pad_type_0"), val = tensor("custom")]; tensor var_9202_pad_0 = const()[name = tensor("op_9202_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696160704)))]; tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696979968)))]; tensor var_9202_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_9200, groups = var_8627, pad = var_9202_pad_0, pad_type = var_9202_pad_type_0, strides = var_9198, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_385_cast_fp16)[name = tensor("op_9202_cast_fp16")]; tensor inputs_65_cast_fp16 = add(x = var_9202_cast_fp16, y = inputs_63_cast_fp16)[name = tensor("inputs_65_cast_fp16")]; tensor var_9206 = const()[name = tensor("op_9206"), val = tensor([1])]; tensor channels_mean_65_cast_fp16 = reduce_mean(axes = var_9206, keep_dims = var_8622, x = inputs_65_cast_fp16)[name = tensor("channels_mean_65_cast_fp16")]; tensor zero_mean_65_cast_fp16 = sub(x = inputs_65_cast_fp16, y = channels_mean_65_cast_fp16)[name = tensor("zero_mean_65_cast_fp16")]; tensor zero_mean_sq_65_cast_fp16 = mul(x = zero_mean_65_cast_fp16, y = zero_mean_65_cast_fp16)[name = tensor("zero_mean_sq_65_cast_fp16")]; tensor var_9210 = const()[name = tensor("op_9210"), val = tensor([1])]; tensor var_9211_cast_fp16 = reduce_mean(axes = var_9210, keep_dims = var_8622, x = zero_mean_sq_65_cast_fp16)[name = tensor("op_9211_cast_fp16")]; tensor var_9212_to_fp16 = const()[name = tensor("op_9212_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_9213_cast_fp16 = add(x = var_9211_cast_fp16, y = var_9212_to_fp16)[name = tensor("op_9213_cast_fp16")]; tensor denom_65_epsilon_0_to_fp16 = const()[name = tensor("denom_65_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_65_cast_fp16 = rsqrt(epsilon = denom_65_epsilon_0_to_fp16, x = var_9213_cast_fp16)[name = tensor("denom_65_cast_fp16")]; tensor out_65_cast_fp16 = mul(x = zero_mean_65_cast_fp16, y = denom_65_cast_fp16)[name = tensor("out_65_cast_fp16")]; tensor var_9217_to_fp16 = const()[name = tensor("op_9217_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696981312)))]; tensor var_9218_cast_fp16 = add(x = out_65_cast_fp16, y = var_9217_to_fp16)[name = tensor("op_9218_cast_fp16")]; tensor var_9220_to_fp16 = const()[name = tensor("op_9220_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696982656)))]; tensor input_387_cast_fp16 = mul(x = var_9218_cast_fp16, y = var_9220_to_fp16)[name = tensor("input_387_cast_fp16")]; tensor var_9228 = const()[name = tensor("op_9228"), val = tensor([1, 1])]; tensor var_9230 = const()[name = tensor("op_9230"), val = tensor([1, 1])]; tensor var_9232_pad_type_0 = const()[name = tensor("op_9232_pad_type_0"), val = tensor("custom")]; tensor var_9232_pad_0 = const()[name = tensor("op_9232_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696984000)))]; tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(703537664)))]; tensor var_9232_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_9230, groups = var_8627, pad = var_9232_pad_0, pad_type = var_9232_pad_type_0, strides = var_9228, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_387_cast_fp16)[name = tensor("op_9232_cast_fp16")]; tensor var_9233_split_sizes_0 = const()[name = tensor("op_9233_split_sizes_0"), val = tensor([2560, 2560])]; tensor var_9233_axis_0 = const()[name = tensor("op_9233_axis_0"), val = tensor(1)]; tensor var_9233_cast_fp16_0, tensor var_9233_cast_fp16_1 = split(axis = var_9233_axis_0, split_sizes = var_9233_split_sizes_0, x = var_9232_cast_fp16)[name = tensor("op_9233_cast_fp16")]; tensor var_9235_mode_0 = const()[name = tensor("op_9235_mode_0"), val = tensor("EXACT")]; tensor var_9235_cast_fp16 = gelu(mode = var_9235_mode_0, x = var_9233_cast_fp16_1)[name = tensor("op_9235_cast_fp16")]; tensor input_389_cast_fp16 = mul(x = var_9233_cast_fp16_0, y = var_9235_cast_fp16)[name = tensor("input_389_cast_fp16")]; tensor var_9239 = const()[name = tensor("op_9239"), val = tensor([1, 1])]; tensor var_9241 = const()[name = tensor("op_9241"), val = tensor([1, 1])]; tensor var_9243_pad_type_0 = const()[name = tensor("op_9243_pad_type_0"), val = tensor("custom")]; tensor var_9243_pad_0 = const()[name = tensor("op_9243_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(703547968)))]; tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(706824832)))]; tensor var_9243_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_9241, groups = var_8627, pad = var_9243_pad_0, pad_type = var_9243_pad_type_0, strides = var_9239, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_389_cast_fp16)[name = tensor("op_9243_cast_fp16")]; tensor hidden_states_233_cast_fp16 = add(x = var_9243_cast_fp16, y = inputs_65_cast_fp16)[name = tensor("hidden_states_233_cast_fp16")]; tensor var_9245 = const()[name = tensor("op_9245"), val = tensor([2, 640, 24, 24])]; tensor input_391_cast_fp16 = reshape(shape = var_9245, x = hidden_states_233_cast_fp16)[name = tensor("input_391_cast_fp16")]; tensor var_9249 = const()[name = tensor("op_9249"), val = tensor([1, 1])]; tensor var_9251 = const()[name = tensor("op_9251"), val = tensor([1, 1])]; tensor hidden_states_235_pad_type_0 = const()[name = tensor("hidden_states_235_pad_type_0"), val = tensor("custom")]; tensor hidden_states_235_pad_0 = const()[name = tensor("hidden_states_235_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(706826176)))]; tensor up_blocks_2_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(707645440)))]; tensor hidden_states_235_cast_fp16 = conv(bias = up_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_9251, groups = var_8627, pad = hidden_states_235_pad_0, pad_type = hidden_states_235_pad_type_0, strides = var_9249, weight = up_blocks_2_attentions_0_proj_out_weight_to_fp16, x = input_391_cast_fp16)[name = tensor("hidden_states_235_cast_fp16")]; tensor hidden_states_237_cast_fp16 = add(x = hidden_states_235_cast_fp16, y = hidden_states_223_cast_fp16)[name = tensor("hidden_states_237_cast_fp16")]; tensor input_393_interleave_0 = const()[name = tensor("input_393_interleave_0"), val = tensor(false)]; tensor cast_8 = cast(dtype = cast_10_dtype_0, x = input_89_cast_fp16)[name = tensor("cast_8")]; tensor input_393_cast_fp16 = concat(axis = var_8627, interleave = input_393_interleave_0, values = (hidden_states_237_cast_fp16, cast_8))[name = tensor("input_393_cast_fp16")]; tensor reshape_180_shape_0 = const()[name = tensor("reshape_180_shape_0"), val = tensor([2, 32, 40, 24, 24])]; tensor reshape_180_cast_fp16 = reshape(shape = reshape_180_shape_0, x = input_393_cast_fp16)[name = tensor("reshape_180_cast_fp16")]; tensor reduce_mean_135_axes_0 = const()[name = tensor("reduce_mean_135_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_135_keep_dims_0 = const()[name = tensor("reduce_mean_135_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_135_cast_fp16 = reduce_mean(axes = reduce_mean_135_axes_0, keep_dims = reduce_mean_135_keep_dims_0, x = reshape_180_cast_fp16)[name = tensor("reduce_mean_135_cast_fp16")]; tensor sub_90_cast_fp16 = sub(x = reshape_180_cast_fp16, y = reduce_mean_135_cast_fp16)[name = tensor("sub_90_cast_fp16")]; tensor square_45_cast_fp16 = square(x = sub_90_cast_fp16)[name = tensor("square_45_cast_fp16")]; tensor reduce_mean_137_axes_0 = const()[name = tensor("reduce_mean_137_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_137_keep_dims_0 = const()[name = tensor("reduce_mean_137_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_137_cast_fp16 = reduce_mean(axes = reduce_mean_137_axes_0, keep_dims = reduce_mean_137_keep_dims_0, x = square_45_cast_fp16)[name = tensor("reduce_mean_137_cast_fp16")]; tensor add_90_y_0_to_fp16 = const()[name = tensor("add_90_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_90_cast_fp16 = add(x = reduce_mean_137_cast_fp16, y = add_90_y_0_to_fp16)[name = tensor("add_90_cast_fp16")]; tensor sqrt_45_cast_fp16 = sqrt(x = add_90_cast_fp16)[name = tensor("sqrt_45_cast_fp16")]; tensor real_div_45_cast_fp16 = real_div(x = sub_90_cast_fp16, y = sqrt_45_cast_fp16)[name = tensor("real_div_45_cast_fp16")]; tensor reshape_181_shape_0 = const()[name = tensor("reshape_181_shape_0"), val = tensor([2, 1280, 24, 24])]; tensor reshape_181_cast_fp16 = reshape(shape = reshape_181_shape_0, x = real_div_45_cast_fp16)[name = tensor("reshape_181_cast_fp16")]; tensor add_91_gamma_0_to_fp16 = const()[name = tensor("add_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(707646784)))]; tensor add_91_beta_0_to_fp16 = const()[name = tensor("add_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(707649408)))]; tensor add_91_epsilon_0_to_fp16 = const()[name = tensor("add_91_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_91_cast_fp16 = batch_norm(beta = add_91_beta_0_to_fp16, epsilon = add_91_epsilon_0_to_fp16, gamma = add_91_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_181_cast_fp16)[name = tensor("add_91_cast_fp16")]; tensor input_397_cast_fp16 = silu(x = add_91_cast_fp16)[name = tensor("input_397_cast_fp16")]; tensor var_9269 = const()[name = tensor("op_9269"), val = tensor([1, 1])]; tensor var_9271 = const()[name = tensor("op_9271"), val = tensor([1, 1])]; tensor hidden_states_239_pad_type_0 = const()[name = tensor("hidden_states_239_pad_type_0"), val = tensor("custom")]; tensor hidden_states_239_pad_0 = const()[name = tensor("hidden_states_239_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_2_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(707652032)))]; tensor up_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722397696)))]; tensor hidden_states_239_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_9271, groups = var_8627, pad = hidden_states_239_pad_0, pad_type = hidden_states_239_pad_type_0, strides = var_9269, weight = up_blocks_2_resnets_1_conv1_weight_to_fp16, x = input_397_cast_fp16)[name = tensor("hidden_states_239_cast_fp16")]; tensor var_9277 = const()[name = tensor("op_9277"), val = tensor([1, 1])]; tensor var_9279 = const()[name = tensor("op_9279"), val = tensor([1, 1])]; tensor temb_35_pad_type_0 = const()[name = tensor("temb_35_pad_type_0"), val = tensor("custom")]; tensor temb_35_pad_0 = const()[name = tensor("temb_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722399040)))]; tensor up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(724037504)))]; tensor temb_35_cast_fp16 = conv(bias = up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_9279, groups = var_8627, pad = temb_35_pad_0, pad_type = temb_35_pad_type_0, strides = var_9277, weight = up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16, x = cast_12)[name = tensor("temb_35_cast_fp16")]; tensor input_401_cast_fp16 = add(x = hidden_states_239_cast_fp16, y = temb_35_cast_fp16)[name = tensor("input_401_cast_fp16")]; tensor reshape_184_shape_0 = const()[name = tensor("reshape_184_shape_0"), val = tensor([2, 32, 20, 24, 24])]; tensor reshape_184_cast_fp16 = reshape(shape = reshape_184_shape_0, x = input_401_cast_fp16)[name = tensor("reshape_184_cast_fp16")]; tensor reduce_mean_138_axes_0 = const()[name = tensor("reduce_mean_138_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_138_keep_dims_0 = const()[name = tensor("reduce_mean_138_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_138_cast_fp16 = reduce_mean(axes = reduce_mean_138_axes_0, keep_dims = reduce_mean_138_keep_dims_0, x = reshape_184_cast_fp16)[name = tensor("reduce_mean_138_cast_fp16")]; tensor sub_92_cast_fp16 = sub(x = reshape_184_cast_fp16, y = reduce_mean_138_cast_fp16)[name = tensor("sub_92_cast_fp16")]; tensor square_46_cast_fp16 = square(x = sub_92_cast_fp16)[name = tensor("square_46_cast_fp16")]; tensor reduce_mean_140_axes_0 = const()[name = tensor("reduce_mean_140_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_140_keep_dims_0 = const()[name = tensor("reduce_mean_140_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_140_cast_fp16 = reduce_mean(axes = reduce_mean_140_axes_0, keep_dims = reduce_mean_140_keep_dims_0, x = square_46_cast_fp16)[name = tensor("reduce_mean_140_cast_fp16")]; tensor add_92_y_0_to_fp16 = const()[name = tensor("add_92_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_92_cast_fp16 = add(x = reduce_mean_140_cast_fp16, y = add_92_y_0_to_fp16)[name = tensor("add_92_cast_fp16")]; tensor sqrt_46_cast_fp16 = sqrt(x = add_92_cast_fp16)[name = tensor("sqrt_46_cast_fp16")]; tensor real_div_46_cast_fp16 = real_div(x = sub_92_cast_fp16, y = sqrt_46_cast_fp16)[name = tensor("real_div_46_cast_fp16")]; tensor reshape_185_shape_0 = const()[name = tensor("reshape_185_shape_0"), val = tensor([2, 640, 24, 24])]; tensor reshape_185_cast_fp16 = reshape(shape = reshape_185_shape_0, x = real_div_46_cast_fp16)[name = tensor("reshape_185_cast_fp16")]; tensor add_93_gamma_0_to_fp16 = const()[name = tensor("add_93_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(724038848)))]; tensor add_93_beta_0_to_fp16 = const()[name = tensor("add_93_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(724040192)))]; tensor add_93_epsilon_0_to_fp16 = const()[name = tensor("add_93_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_93_cast_fp16 = batch_norm(beta = add_93_beta_0_to_fp16, epsilon = add_93_epsilon_0_to_fp16, gamma = add_93_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_185_cast_fp16)[name = tensor("add_93_cast_fp16")]; tensor input_405_cast_fp16 = silu(x = add_93_cast_fp16)[name = tensor("input_405_cast_fp16")]; tensor var_9289 = const()[name = tensor("op_9289"), val = tensor([1, 1])]; tensor var_9291 = const()[name = tensor("op_9291"), val = tensor([1, 1])]; tensor hidden_states_241_pad_type_0 = const()[name = tensor("hidden_states_241_pad_type_0"), val = tensor("custom")]; tensor hidden_states_241_pad_0 = const()[name = tensor("hidden_states_241_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_2_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(724041536)))]; tensor up_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(731414400)))]; tensor hidden_states_241_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_9291, groups = var_8627, pad = hidden_states_241_pad_0, pad_type = hidden_states_241_pad_type_0, strides = var_9289, weight = up_blocks_2_resnets_1_conv2_weight_to_fp16, x = input_405_cast_fp16)[name = tensor("hidden_states_241_cast_fp16")]; tensor var_9296 = const()[name = tensor("op_9296"), val = tensor([1, 1])]; tensor var_9298 = const()[name = tensor("op_9298"), val = tensor([1, 1])]; tensor x_19_pad_type_0 = const()[name = tensor("x_19_pad_type_0"), val = tensor("custom")]; tensor x_19_pad_0 = const()[name = tensor("x_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(731415744)))]; tensor up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(733054208)))]; tensor x_19_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_9298, groups = var_8627, pad = x_19_pad_0, pad_type = x_19_pad_type_0, strides = var_9296, weight = up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16, x = input_393_cast_fp16)[name = tensor("x_19_cast_fp16")]; tensor hidden_states_243_cast_fp16 = add(x = x_19_cast_fp16, y = hidden_states_241_cast_fp16)[name = tensor("hidden_states_243_cast_fp16")]; tensor reshape_188_shape_0 = const()[name = tensor("reshape_188_shape_0"), val = tensor([2, 32, 20, 24, 24])]; tensor reshape_188_cast_fp16 = reshape(shape = reshape_188_shape_0, x = hidden_states_243_cast_fp16)[name = tensor("reshape_188_cast_fp16")]; tensor reduce_mean_141_axes_0 = const()[name = tensor("reduce_mean_141_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_141_keep_dims_0 = const()[name = tensor("reduce_mean_141_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_141_cast_fp16 = reduce_mean(axes = reduce_mean_141_axes_0, keep_dims = reduce_mean_141_keep_dims_0, x = reshape_188_cast_fp16)[name = tensor("reduce_mean_141_cast_fp16")]; tensor sub_94_cast_fp16 = sub(x = reshape_188_cast_fp16, y = reduce_mean_141_cast_fp16)[name = tensor("sub_94_cast_fp16")]; tensor square_47_cast_fp16 = square(x = sub_94_cast_fp16)[name = tensor("square_47_cast_fp16")]; tensor reduce_mean_143_axes_0 = const()[name = tensor("reduce_mean_143_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_143_keep_dims_0 = const()[name = tensor("reduce_mean_143_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_143_cast_fp16 = reduce_mean(axes = reduce_mean_143_axes_0, keep_dims = reduce_mean_143_keep_dims_0, x = square_47_cast_fp16)[name = tensor("reduce_mean_143_cast_fp16")]; tensor add_94_y_0_to_fp16 = const()[name = tensor("add_94_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_94_cast_fp16 = add(x = reduce_mean_143_cast_fp16, y = add_94_y_0_to_fp16)[name = tensor("add_94_cast_fp16")]; tensor sqrt_47_cast_fp16 = sqrt(x = add_94_cast_fp16)[name = tensor("sqrt_47_cast_fp16")]; tensor real_div_47_cast_fp16 = real_div(x = sub_94_cast_fp16, y = sqrt_47_cast_fp16)[name = tensor("real_div_47_cast_fp16")]; tensor reshape_189_shape_0 = const()[name = tensor("reshape_189_shape_0"), val = tensor([2, 640, 24, 24])]; tensor reshape_189_cast_fp16 = reshape(shape = reshape_189_shape_0, x = real_div_47_cast_fp16)[name = tensor("reshape_189_cast_fp16")]; tensor add_95_gamma_0_to_fp16 = const()[name = tensor("add_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(733055552)))]; tensor add_95_beta_0_to_fp16 = const()[name = tensor("add_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(733056896)))]; tensor add_95_epsilon_0_to_fp16 = const()[name = tensor("add_95_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_95_cast_fp16 = batch_norm(beta = add_95_beta_0_to_fp16, epsilon = add_95_epsilon_0_to_fp16, gamma = add_95_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_189_cast_fp16)[name = tensor("add_95_cast_fp16")]; tensor var_9318 = const()[name = tensor("op_9318"), val = tensor([1, 1])]; tensor var_9320 = const()[name = tensor("op_9320"), val = tensor([1, 1])]; tensor hidden_states_245_pad_type_0 = const()[name = tensor("hidden_states_245_pad_type_0"), val = tensor("custom")]; tensor hidden_states_245_pad_0 = const()[name = tensor("hidden_states_245_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(733058240)))]; tensor up_blocks_2_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(733877504)))]; tensor hidden_states_245_cast_fp16 = conv(bias = up_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_9320, groups = var_8627, pad = hidden_states_245_pad_0, pad_type = hidden_states_245_pad_type_0, strides = var_9318, weight = up_blocks_2_attentions_1_proj_in_weight_to_fp16, x = add_95_cast_fp16)[name = tensor("hidden_states_245_cast_fp16")]; tensor var_9325 = const()[name = tensor("op_9325"), val = tensor([2, 640, 1, 576])]; tensor inputs_67_cast_fp16 = reshape(shape = var_9325, x = hidden_states_245_cast_fp16)[name = tensor("inputs_67_cast_fp16")]; tensor var_9335 = const()[name = tensor("op_9335"), val = tensor([1])]; tensor channels_mean_67_cast_fp16 = reduce_mean(axes = var_9335, keep_dims = var_8622, x = inputs_67_cast_fp16)[name = tensor("channels_mean_67_cast_fp16")]; tensor zero_mean_67_cast_fp16 = sub(x = inputs_67_cast_fp16, y = channels_mean_67_cast_fp16)[name = tensor("zero_mean_67_cast_fp16")]; tensor zero_mean_sq_67_cast_fp16 = mul(x = zero_mean_67_cast_fp16, y = zero_mean_67_cast_fp16)[name = tensor("zero_mean_sq_67_cast_fp16")]; tensor var_9339 = const()[name = tensor("op_9339"), val = tensor([1])]; tensor var_9340_cast_fp16 = reduce_mean(axes = var_9339, keep_dims = var_8622, x = zero_mean_sq_67_cast_fp16)[name = tensor("op_9340_cast_fp16")]; tensor var_9341_to_fp16 = const()[name = tensor("op_9341_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_9342_cast_fp16 = add(x = var_9340_cast_fp16, y = var_9341_to_fp16)[name = tensor("op_9342_cast_fp16")]; tensor denom_67_epsilon_0_to_fp16 = const()[name = tensor("denom_67_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_67_cast_fp16 = rsqrt(epsilon = denom_67_epsilon_0_to_fp16, x = var_9342_cast_fp16)[name = tensor("denom_67_cast_fp16")]; tensor out_67_cast_fp16 = mul(x = zero_mean_67_cast_fp16, y = denom_67_cast_fp16)[name = tensor("out_67_cast_fp16")]; tensor var_9346_to_fp16 = const()[name = tensor("op_9346_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(733878848)))]; tensor var_9347_cast_fp16 = add(x = out_67_cast_fp16, y = var_9346_to_fp16)[name = tensor("op_9347_cast_fp16")]; tensor var_9349_to_fp16 = const()[name = tensor("op_9349_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(733880192)))]; tensor hidden_states_247_cast_fp16 = mul(x = var_9347_cast_fp16, y = var_9349_to_fp16)[name = tensor("hidden_states_247_cast_fp16")]; tensor var_9356 = const()[name = tensor("op_9356"), val = tensor([1, 1])]; tensor var_9358 = const()[name = tensor("op_9358"), val = tensor([1, 1])]; tensor q_45_pad_type_0 = const()[name = tensor("q_45_pad_type_0"), val = tensor("custom")]; tensor q_45_pad_0 = const()[name = tensor("q_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(733881536)))]; tensor q_45_cast_fp16 = conv(dilations = var_9358, groups = var_8627, pad = q_45_pad_0, pad_type = q_45_pad_type_0, strides = var_9356, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_247_cast_fp16)[name = tensor("q_45_cast_fp16")]; tensor var_9362 = const()[name = tensor("op_9362"), val = tensor([1, 1])]; tensor var_9364 = const()[name = tensor("op_9364"), val = tensor([1, 1])]; tensor k_89_pad_type_0 = const()[name = tensor("k_89_pad_type_0"), val = tensor("custom")]; tensor k_89_pad_0 = const()[name = tensor("k_89_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(734700800)))]; tensor k_89_cast_fp16 = conv(dilations = var_9364, groups = var_8627, pad = k_89_pad_0, pad_type = k_89_pad_type_0, strides = var_9362, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_247_cast_fp16)[name = tensor("k_89_cast_fp16")]; tensor var_9368 = const()[name = tensor("op_9368"), val = tensor([1, 1])]; tensor var_9370 = const()[name = tensor("op_9370"), val = tensor([1, 1])]; tensor v_45_pad_type_0 = const()[name = tensor("v_45_pad_type_0"), val = tensor("custom")]; tensor v_45_pad_0 = const()[name = tensor("v_45_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735520064)))]; tensor v_45_cast_fp16 = conv(dilations = var_9370, groups = var_8627, pad = v_45_pad_0, pad_type = v_45_pad_type_0, strides = var_9368, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_247_cast_fp16)[name = tensor("v_45_cast_fp16")]; tensor var_9374_begin_0 = const()[name = tensor("op_9374_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9374_end_0 = const()[name = tensor("op_9374_end_0"), val = tensor([2, 64, 1, 576])]; tensor var_9374_end_mask_0 = const()[name = tensor("op_9374_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9374_cast_fp16 = slice_by_index(begin = var_9374_begin_0, end = var_9374_end_0, end_mask = var_9374_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9374_cast_fp16")]; tensor var_9378_begin_0 = const()[name = tensor("op_9378_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_9378_end_0 = const()[name = tensor("op_9378_end_0"), val = tensor([2, 128, 1, 576])]; tensor var_9378_end_mask_0 = const()[name = tensor("op_9378_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9378_cast_fp16 = slice_by_index(begin = var_9378_begin_0, end = var_9378_end_0, end_mask = var_9378_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9378_cast_fp16")]; tensor var_9382_begin_0 = const()[name = tensor("op_9382_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_9382_end_0 = const()[name = tensor("op_9382_end_0"), val = tensor([2, 192, 1, 576])]; tensor var_9382_end_mask_0 = const()[name = tensor("op_9382_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9382_cast_fp16 = slice_by_index(begin = var_9382_begin_0, end = var_9382_end_0, end_mask = var_9382_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9382_cast_fp16")]; tensor var_9386_begin_0 = const()[name = tensor("op_9386_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_9386_end_0 = const()[name = tensor("op_9386_end_0"), val = tensor([2, 256, 1, 576])]; tensor var_9386_end_mask_0 = const()[name = tensor("op_9386_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9386_cast_fp16 = slice_by_index(begin = var_9386_begin_0, end = var_9386_end_0, end_mask = var_9386_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9386_cast_fp16")]; tensor var_9390_begin_0 = const()[name = tensor("op_9390_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_9390_end_0 = const()[name = tensor("op_9390_end_0"), val = tensor([2, 320, 1, 576])]; tensor var_9390_end_mask_0 = const()[name = tensor("op_9390_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9390_cast_fp16 = slice_by_index(begin = var_9390_begin_0, end = var_9390_end_0, end_mask = var_9390_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9390_cast_fp16")]; tensor var_9394_begin_0 = const()[name = tensor("op_9394_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9394_end_0 = const()[name = tensor("op_9394_end_0"), val = tensor([2, 384, 1, 576])]; tensor var_9394_end_mask_0 = const()[name = tensor("op_9394_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9394_cast_fp16 = slice_by_index(begin = var_9394_begin_0, end = var_9394_end_0, end_mask = var_9394_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9394_cast_fp16")]; tensor var_9398_begin_0 = const()[name = tensor("op_9398_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_9398_end_0 = const()[name = tensor("op_9398_end_0"), val = tensor([2, 448, 1, 576])]; tensor var_9398_end_mask_0 = const()[name = tensor("op_9398_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9398_cast_fp16 = slice_by_index(begin = var_9398_begin_0, end = var_9398_end_0, end_mask = var_9398_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9398_cast_fp16")]; tensor var_9402_begin_0 = const()[name = tensor("op_9402_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_9402_end_0 = const()[name = tensor("op_9402_end_0"), val = tensor([2, 512, 1, 576])]; tensor var_9402_end_mask_0 = const()[name = tensor("op_9402_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9402_cast_fp16 = slice_by_index(begin = var_9402_begin_0, end = var_9402_end_0, end_mask = var_9402_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9402_cast_fp16")]; tensor var_9406_begin_0 = const()[name = tensor("op_9406_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_9406_end_0 = const()[name = tensor("op_9406_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_9406_end_mask_0 = const()[name = tensor("op_9406_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9406_cast_fp16 = slice_by_index(begin = var_9406_begin_0, end = var_9406_end_0, end_mask = var_9406_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9406_cast_fp16")]; tensor var_9410_begin_0 = const()[name = tensor("op_9410_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_9410_end_0 = const()[name = tensor("op_9410_end_0"), val = tensor([2, 640, 1, 576])]; tensor var_9410_end_mask_0 = const()[name = tensor("op_9410_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9410_cast_fp16 = slice_by_index(begin = var_9410_begin_0, end = var_9410_end_0, end_mask = var_9410_end_mask_0, x = q_45_cast_fp16)[name = tensor("op_9410_cast_fp16")]; tensor k_91_perm_0 = const()[name = tensor("k_91_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_9417_begin_0 = const()[name = tensor("op_9417_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9417_end_0 = const()[name = tensor("op_9417_end_0"), val = tensor([2, 576, 1, 64])]; tensor var_9417_end_mask_0 = const()[name = tensor("op_9417_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_9 = transpose(perm = k_91_perm_0, x = k_89_cast_fp16)[name = tensor("transpose_9")]; tensor var_9417_cast_fp16 = slice_by_index(begin = var_9417_begin_0, end = var_9417_end_0, end_mask = var_9417_end_mask_0, x = transpose_9)[name = tensor("op_9417_cast_fp16")]; tensor var_9421_begin_0 = const()[name = tensor("op_9421_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_9421_end_0 = const()[name = tensor("op_9421_end_0"), val = tensor([2, 576, 1, 128])]; tensor var_9421_end_mask_0 = const()[name = tensor("op_9421_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9421_cast_fp16 = slice_by_index(begin = var_9421_begin_0, end = var_9421_end_0, end_mask = var_9421_end_mask_0, x = transpose_9)[name = tensor("op_9421_cast_fp16")]; tensor var_9425_begin_0 = const()[name = tensor("op_9425_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_9425_end_0 = const()[name = tensor("op_9425_end_0"), val = tensor([2, 576, 1, 192])]; tensor var_9425_end_mask_0 = const()[name = tensor("op_9425_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9425_cast_fp16 = slice_by_index(begin = var_9425_begin_0, end = var_9425_end_0, end_mask = var_9425_end_mask_0, x = transpose_9)[name = tensor("op_9425_cast_fp16")]; tensor var_9429_begin_0 = const()[name = tensor("op_9429_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_9429_end_0 = const()[name = tensor("op_9429_end_0"), val = tensor([2, 576, 1, 256])]; tensor var_9429_end_mask_0 = const()[name = tensor("op_9429_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9429_cast_fp16 = slice_by_index(begin = var_9429_begin_0, end = var_9429_end_0, end_mask = var_9429_end_mask_0, x = transpose_9)[name = tensor("op_9429_cast_fp16")]; tensor var_9433_begin_0 = const()[name = tensor("op_9433_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_9433_end_0 = const()[name = tensor("op_9433_end_0"), val = tensor([2, 576, 1, 320])]; tensor var_9433_end_mask_0 = const()[name = tensor("op_9433_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9433_cast_fp16 = slice_by_index(begin = var_9433_begin_0, end = var_9433_end_0, end_mask = var_9433_end_mask_0, x = transpose_9)[name = tensor("op_9433_cast_fp16")]; tensor var_9437_begin_0 = const()[name = tensor("op_9437_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_9437_end_0 = const()[name = tensor("op_9437_end_0"), val = tensor([2, 576, 1, 384])]; tensor var_9437_end_mask_0 = const()[name = tensor("op_9437_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9437_cast_fp16 = slice_by_index(begin = var_9437_begin_0, end = var_9437_end_0, end_mask = var_9437_end_mask_0, x = transpose_9)[name = tensor("op_9437_cast_fp16")]; tensor var_9441_begin_0 = const()[name = tensor("op_9441_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_9441_end_0 = const()[name = tensor("op_9441_end_0"), val = tensor([2, 576, 1, 448])]; tensor var_9441_end_mask_0 = const()[name = tensor("op_9441_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9441_cast_fp16 = slice_by_index(begin = var_9441_begin_0, end = var_9441_end_0, end_mask = var_9441_end_mask_0, x = transpose_9)[name = tensor("op_9441_cast_fp16")]; tensor var_9445_begin_0 = const()[name = tensor("op_9445_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_9445_end_0 = const()[name = tensor("op_9445_end_0"), val = tensor([2, 576, 1, 512])]; tensor var_9445_end_mask_0 = const()[name = tensor("op_9445_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9445_cast_fp16 = slice_by_index(begin = var_9445_begin_0, end = var_9445_end_0, end_mask = var_9445_end_mask_0, x = transpose_9)[name = tensor("op_9445_cast_fp16")]; tensor var_9449_begin_0 = const()[name = tensor("op_9449_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9449_end_0 = const()[name = tensor("op_9449_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_9449_end_mask_0 = const()[name = tensor("op_9449_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9449_cast_fp16 = slice_by_index(begin = var_9449_begin_0, end = var_9449_end_0, end_mask = var_9449_end_mask_0, x = transpose_9)[name = tensor("op_9449_cast_fp16")]; tensor var_9453_begin_0 = const()[name = tensor("op_9453_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_9453_end_0 = const()[name = tensor("op_9453_end_0"), val = tensor([2, 576, 1, 640])]; tensor var_9453_end_mask_0 = const()[name = tensor("op_9453_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9453_cast_fp16 = slice_by_index(begin = var_9453_begin_0, end = var_9453_end_0, end_mask = var_9453_end_mask_0, x = transpose_9)[name = tensor("op_9453_cast_fp16")]; tensor var_9455_begin_0 = const()[name = tensor("op_9455_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9455_end_0 = const()[name = tensor("op_9455_end_0"), val = tensor([2, 64, 1, 576])]; tensor var_9455_end_mask_0 = const()[name = tensor("op_9455_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9455_cast_fp16 = slice_by_index(begin = var_9455_begin_0, end = var_9455_end_0, end_mask = var_9455_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9455_cast_fp16")]; tensor var_9459_begin_0 = const()[name = tensor("op_9459_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_9459_end_0 = const()[name = tensor("op_9459_end_0"), val = tensor([2, 128, 1, 576])]; tensor var_9459_end_mask_0 = const()[name = tensor("op_9459_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9459_cast_fp16 = slice_by_index(begin = var_9459_begin_0, end = var_9459_end_0, end_mask = var_9459_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9459_cast_fp16")]; tensor var_9463_begin_0 = const()[name = tensor("op_9463_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_9463_end_0 = const()[name = tensor("op_9463_end_0"), val = tensor([2, 192, 1, 576])]; tensor var_9463_end_mask_0 = const()[name = tensor("op_9463_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9463_cast_fp16 = slice_by_index(begin = var_9463_begin_0, end = var_9463_end_0, end_mask = var_9463_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9463_cast_fp16")]; tensor var_9467_begin_0 = const()[name = tensor("op_9467_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_9467_end_0 = const()[name = tensor("op_9467_end_0"), val = tensor([2, 256, 1, 576])]; tensor var_9467_end_mask_0 = const()[name = tensor("op_9467_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9467_cast_fp16 = slice_by_index(begin = var_9467_begin_0, end = var_9467_end_0, end_mask = var_9467_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9467_cast_fp16")]; tensor var_9471_begin_0 = const()[name = tensor("op_9471_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_9471_end_0 = const()[name = tensor("op_9471_end_0"), val = tensor([2, 320, 1, 576])]; tensor var_9471_end_mask_0 = const()[name = tensor("op_9471_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9471_cast_fp16 = slice_by_index(begin = var_9471_begin_0, end = var_9471_end_0, end_mask = var_9471_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9471_cast_fp16")]; tensor var_9475_begin_0 = const()[name = tensor("op_9475_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9475_end_0 = const()[name = tensor("op_9475_end_0"), val = tensor([2, 384, 1, 576])]; tensor var_9475_end_mask_0 = const()[name = tensor("op_9475_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9475_cast_fp16 = slice_by_index(begin = var_9475_begin_0, end = var_9475_end_0, end_mask = var_9475_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9475_cast_fp16")]; tensor var_9479_begin_0 = const()[name = tensor("op_9479_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_9479_end_0 = const()[name = tensor("op_9479_end_0"), val = tensor([2, 448, 1, 576])]; tensor var_9479_end_mask_0 = const()[name = tensor("op_9479_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9479_cast_fp16 = slice_by_index(begin = var_9479_begin_0, end = var_9479_end_0, end_mask = var_9479_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9479_cast_fp16")]; tensor var_9483_begin_0 = const()[name = tensor("op_9483_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_9483_end_0 = const()[name = tensor("op_9483_end_0"), val = tensor([2, 512, 1, 576])]; tensor var_9483_end_mask_0 = const()[name = tensor("op_9483_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9483_cast_fp16 = slice_by_index(begin = var_9483_begin_0, end = var_9483_end_0, end_mask = var_9483_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9483_cast_fp16")]; tensor var_9487_begin_0 = const()[name = tensor("op_9487_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_9487_end_0 = const()[name = tensor("op_9487_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_9487_end_mask_0 = const()[name = tensor("op_9487_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9487_cast_fp16 = slice_by_index(begin = var_9487_begin_0, end = var_9487_end_0, end_mask = var_9487_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9487_cast_fp16")]; tensor var_9491_begin_0 = const()[name = tensor("op_9491_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_9491_end_0 = const()[name = tensor("op_9491_end_0"), val = tensor([2, 640, 1, 576])]; tensor var_9491_end_mask_0 = const()[name = tensor("op_9491_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9491_cast_fp16 = slice_by_index(begin = var_9491_begin_0, end = var_9491_end_0, end_mask = var_9491_end_mask_0, x = v_45_cast_fp16)[name = tensor("op_9491_cast_fp16")]; tensor var_9495_equation_0 = const()[name = tensor("op_9495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9495_cast_fp16 = einsum(equation = var_9495_equation_0, values = (var_9417_cast_fp16, var_9374_cast_fp16))[name = tensor("op_9495_cast_fp16")]; tensor var_9496_to_fp16 = const()[name = tensor("op_9496_to_fp16"), val = tensor(0x1p-3)]; tensor aw_641_cast_fp16 = mul(x = var_9495_cast_fp16, y = var_9496_to_fp16)[name = tensor("aw_641_cast_fp16")]; tensor var_9499_equation_0 = const()[name = tensor("op_9499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9499_cast_fp16 = einsum(equation = var_9499_equation_0, values = (var_9421_cast_fp16, var_9378_cast_fp16))[name = tensor("op_9499_cast_fp16")]; tensor var_9500_to_fp16 = const()[name = tensor("op_9500_to_fp16"), val = tensor(0x1p-3)]; tensor aw_643_cast_fp16 = mul(x = var_9499_cast_fp16, y = var_9500_to_fp16)[name = tensor("aw_643_cast_fp16")]; tensor var_9503_equation_0 = const()[name = tensor("op_9503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9503_cast_fp16 = einsum(equation = var_9503_equation_0, values = (var_9425_cast_fp16, var_9382_cast_fp16))[name = tensor("op_9503_cast_fp16")]; tensor var_9504_to_fp16 = const()[name = tensor("op_9504_to_fp16"), val = tensor(0x1p-3)]; tensor aw_645_cast_fp16 = mul(x = var_9503_cast_fp16, y = var_9504_to_fp16)[name = tensor("aw_645_cast_fp16")]; tensor var_9507_equation_0 = const()[name = tensor("op_9507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9507_cast_fp16 = einsum(equation = var_9507_equation_0, values = (var_9429_cast_fp16, var_9386_cast_fp16))[name = tensor("op_9507_cast_fp16")]; tensor var_9508_to_fp16 = const()[name = tensor("op_9508_to_fp16"), val = tensor(0x1p-3)]; tensor aw_647_cast_fp16 = mul(x = var_9507_cast_fp16, y = var_9508_to_fp16)[name = tensor("aw_647_cast_fp16")]; tensor var_9511_equation_0 = const()[name = tensor("op_9511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9511_cast_fp16 = einsum(equation = var_9511_equation_0, values = (var_9433_cast_fp16, var_9390_cast_fp16))[name = tensor("op_9511_cast_fp16")]; tensor var_9512_to_fp16 = const()[name = tensor("op_9512_to_fp16"), val = tensor(0x1p-3)]; tensor aw_649_cast_fp16 = mul(x = var_9511_cast_fp16, y = var_9512_to_fp16)[name = tensor("aw_649_cast_fp16")]; tensor var_9515_equation_0 = const()[name = tensor("op_9515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9515_cast_fp16 = einsum(equation = var_9515_equation_0, values = (var_9437_cast_fp16, var_9394_cast_fp16))[name = tensor("op_9515_cast_fp16")]; tensor var_9516_to_fp16 = const()[name = tensor("op_9516_to_fp16"), val = tensor(0x1p-3)]; tensor aw_651_cast_fp16 = mul(x = var_9515_cast_fp16, y = var_9516_to_fp16)[name = tensor("aw_651_cast_fp16")]; tensor var_9519_equation_0 = const()[name = tensor("op_9519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9519_cast_fp16 = einsum(equation = var_9519_equation_0, values = (var_9441_cast_fp16, var_9398_cast_fp16))[name = tensor("op_9519_cast_fp16")]; tensor var_9520_to_fp16 = const()[name = tensor("op_9520_to_fp16"), val = tensor(0x1p-3)]; tensor aw_653_cast_fp16 = mul(x = var_9519_cast_fp16, y = var_9520_to_fp16)[name = tensor("aw_653_cast_fp16")]; tensor var_9523_equation_0 = const()[name = tensor("op_9523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9523_cast_fp16 = einsum(equation = var_9523_equation_0, values = (var_9445_cast_fp16, var_9402_cast_fp16))[name = tensor("op_9523_cast_fp16")]; tensor var_9524_to_fp16 = const()[name = tensor("op_9524_to_fp16"), val = tensor(0x1p-3)]; tensor aw_655_cast_fp16 = mul(x = var_9523_cast_fp16, y = var_9524_to_fp16)[name = tensor("aw_655_cast_fp16")]; tensor var_9527_equation_0 = const()[name = tensor("op_9527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9527_cast_fp16 = einsum(equation = var_9527_equation_0, values = (var_9449_cast_fp16, var_9406_cast_fp16))[name = tensor("op_9527_cast_fp16")]; tensor var_9528_to_fp16 = const()[name = tensor("op_9528_to_fp16"), val = tensor(0x1p-3)]; tensor aw_657_cast_fp16 = mul(x = var_9527_cast_fp16, y = var_9528_to_fp16)[name = tensor("aw_657_cast_fp16")]; tensor var_9531_equation_0 = const()[name = tensor("op_9531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9531_cast_fp16 = einsum(equation = var_9531_equation_0, values = (var_9453_cast_fp16, var_9410_cast_fp16))[name = tensor("op_9531_cast_fp16")]; tensor var_9532_to_fp16 = const()[name = tensor("op_9532_to_fp16"), val = tensor(0x1p-3)]; tensor aw_659_cast_fp16 = mul(x = var_9531_cast_fp16, y = var_9532_to_fp16)[name = tensor("aw_659_cast_fp16")]; tensor var_9534_cast_fp16 = softmax(axis = var_8627, x = aw_641_cast_fp16)[name = tensor("op_9534_cast_fp16")]; tensor var_9535_cast_fp16 = softmax(axis = var_8627, x = aw_643_cast_fp16)[name = tensor("op_9535_cast_fp16")]; tensor var_9536_cast_fp16 = softmax(axis = var_8627, x = aw_645_cast_fp16)[name = tensor("op_9536_cast_fp16")]; tensor var_9537_cast_fp16 = softmax(axis = var_8627, x = aw_647_cast_fp16)[name = tensor("op_9537_cast_fp16")]; tensor var_9538_cast_fp16 = softmax(axis = var_8627, x = aw_649_cast_fp16)[name = tensor("op_9538_cast_fp16")]; tensor var_9539_cast_fp16 = softmax(axis = var_8627, x = aw_651_cast_fp16)[name = tensor("op_9539_cast_fp16")]; tensor var_9540_cast_fp16 = softmax(axis = var_8627, x = aw_653_cast_fp16)[name = tensor("op_9540_cast_fp16")]; tensor var_9541_cast_fp16 = softmax(axis = var_8627, x = aw_655_cast_fp16)[name = tensor("op_9541_cast_fp16")]; tensor var_9542_cast_fp16 = softmax(axis = var_8627, x = aw_657_cast_fp16)[name = tensor("op_9542_cast_fp16")]; tensor var_9543_cast_fp16 = softmax(axis = var_8627, x = aw_659_cast_fp16)[name = tensor("op_9543_cast_fp16")]; tensor var_9545_equation_0 = const()[name = tensor("op_9545_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9545_cast_fp16 = einsum(equation = var_9545_equation_0, values = (var_9455_cast_fp16, var_9534_cast_fp16))[name = tensor("op_9545_cast_fp16")]; tensor var_9547_equation_0 = const()[name = tensor("op_9547_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9547_cast_fp16 = einsum(equation = var_9547_equation_0, values = (var_9459_cast_fp16, var_9535_cast_fp16))[name = tensor("op_9547_cast_fp16")]; tensor var_9549_equation_0 = const()[name = tensor("op_9549_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9549_cast_fp16 = einsum(equation = var_9549_equation_0, values = (var_9463_cast_fp16, var_9536_cast_fp16))[name = tensor("op_9549_cast_fp16")]; tensor var_9551_equation_0 = const()[name = tensor("op_9551_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9551_cast_fp16 = einsum(equation = var_9551_equation_0, values = (var_9467_cast_fp16, var_9537_cast_fp16))[name = tensor("op_9551_cast_fp16")]; tensor var_9553_equation_0 = const()[name = tensor("op_9553_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9553_cast_fp16 = einsum(equation = var_9553_equation_0, values = (var_9471_cast_fp16, var_9538_cast_fp16))[name = tensor("op_9553_cast_fp16")]; tensor var_9555_equation_0 = const()[name = tensor("op_9555_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9555_cast_fp16 = einsum(equation = var_9555_equation_0, values = (var_9475_cast_fp16, var_9539_cast_fp16))[name = tensor("op_9555_cast_fp16")]; tensor var_9557_equation_0 = const()[name = tensor("op_9557_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9557_cast_fp16 = einsum(equation = var_9557_equation_0, values = (var_9479_cast_fp16, var_9540_cast_fp16))[name = tensor("op_9557_cast_fp16")]; tensor var_9559_equation_0 = const()[name = tensor("op_9559_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9559_cast_fp16 = einsum(equation = var_9559_equation_0, values = (var_9483_cast_fp16, var_9541_cast_fp16))[name = tensor("op_9559_cast_fp16")]; tensor var_9561_equation_0 = const()[name = tensor("op_9561_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9561_cast_fp16 = einsum(equation = var_9561_equation_0, values = (var_9487_cast_fp16, var_9542_cast_fp16))[name = tensor("op_9561_cast_fp16")]; tensor var_9563_equation_0 = const()[name = tensor("op_9563_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9563_cast_fp16 = einsum(equation = var_9563_equation_0, values = (var_9491_cast_fp16, var_9543_cast_fp16))[name = tensor("op_9563_cast_fp16")]; tensor input_409_interleave_0 = const()[name = tensor("input_409_interleave_0"), val = tensor(false)]; tensor input_409_cast_fp16 = concat(axis = var_8627, interleave = input_409_interleave_0, values = (var_9545_cast_fp16, var_9547_cast_fp16, var_9549_cast_fp16, var_9551_cast_fp16, var_9553_cast_fp16, var_9555_cast_fp16, var_9557_cast_fp16, var_9559_cast_fp16, var_9561_cast_fp16, var_9563_cast_fp16))[name = tensor("input_409_cast_fp16")]; tensor var_9569 = const()[name = tensor("op_9569"), val = tensor([1, 1])]; tensor var_9571 = const()[name = tensor("op_9571"), val = tensor([1, 1])]; tensor var_9573_pad_type_0 = const()[name = tensor("op_9573_pad_type_0"), val = tensor("custom")]; tensor var_9573_pad_0 = const()[name = tensor("op_9573_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736339328)))]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737158592)))]; tensor var_9573_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_9571, groups = var_8627, pad = var_9573_pad_0, pad_type = var_9573_pad_type_0, strides = var_9569, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_409_cast_fp16)[name = tensor("op_9573_cast_fp16")]; tensor inputs_69_cast_fp16 = add(x = var_9573_cast_fp16, y = inputs_67_cast_fp16)[name = tensor("inputs_69_cast_fp16")]; tensor var_9577 = const()[name = tensor("op_9577"), val = tensor([1])]; tensor channels_mean_69_cast_fp16 = reduce_mean(axes = var_9577, keep_dims = var_8622, x = inputs_69_cast_fp16)[name = tensor("channels_mean_69_cast_fp16")]; tensor zero_mean_69_cast_fp16 = sub(x = inputs_69_cast_fp16, y = channels_mean_69_cast_fp16)[name = tensor("zero_mean_69_cast_fp16")]; tensor zero_mean_sq_69_cast_fp16 = mul(x = zero_mean_69_cast_fp16, y = zero_mean_69_cast_fp16)[name = tensor("zero_mean_sq_69_cast_fp16")]; tensor var_9581 = const()[name = tensor("op_9581"), val = tensor([1])]; tensor var_9582_cast_fp16 = reduce_mean(axes = var_9581, keep_dims = var_8622, x = zero_mean_sq_69_cast_fp16)[name = tensor("op_9582_cast_fp16")]; tensor var_9583_to_fp16 = const()[name = tensor("op_9583_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_9584_cast_fp16 = add(x = var_9582_cast_fp16, y = var_9583_to_fp16)[name = tensor("op_9584_cast_fp16")]; tensor denom_69_epsilon_0_to_fp16 = const()[name = tensor("denom_69_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_69_cast_fp16 = rsqrt(epsilon = denom_69_epsilon_0_to_fp16, x = var_9584_cast_fp16)[name = tensor("denom_69_cast_fp16")]; tensor out_69_cast_fp16 = mul(x = zero_mean_69_cast_fp16, y = denom_69_cast_fp16)[name = tensor("out_69_cast_fp16")]; tensor var_9588_to_fp16 = const()[name = tensor("op_9588_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737159936)))]; tensor var_9589_cast_fp16 = add(x = out_69_cast_fp16, y = var_9588_to_fp16)[name = tensor("op_9589_cast_fp16")]; tensor var_9591_to_fp16 = const()[name = tensor("op_9591_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737161280)))]; tensor hidden_states_249_cast_fp16 = mul(x = var_9589_cast_fp16, y = var_9591_to_fp16)[name = tensor("hidden_states_249_cast_fp16")]; tensor var_9598 = const()[name = tensor("op_9598"), val = tensor([1, 1])]; tensor var_9600 = const()[name = tensor("op_9600"), val = tensor([1, 1])]; tensor q_47_pad_type_0 = const()[name = tensor("q_47_pad_type_0"), val = tensor("custom")]; tensor q_47_pad_0 = const()[name = tensor("q_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737162624)))]; tensor q_47_cast_fp16 = conv(dilations = var_9600, groups = var_8627, pad = q_47_pad_0, pad_type = q_47_pad_type_0, strides = var_9598, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_249_cast_fp16)[name = tensor("q_47_cast_fp16")]; tensor var_9604 = const()[name = tensor("op_9604"), val = tensor([1, 1])]; tensor var_9606 = const()[name = tensor("op_9606"), val = tensor([1, 1])]; tensor k_93_pad_type_0 = const()[name = tensor("k_93_pad_type_0"), val = tensor("custom")]; tensor k_93_pad_0 = const()[name = tensor("k_93_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(737981888)))]; tensor k_93_cast_fp16 = conv(dilations = var_9606, groups = var_8627, pad = k_93_pad_0, pad_type = k_93_pad_type_0, strides = var_9604, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_93_cast_fp16")]; tensor var_9610 = const()[name = tensor("op_9610"), val = tensor([1, 1])]; tensor var_9612 = const()[name = tensor("op_9612"), val = tensor([1, 1])]; tensor v_47_pad_type_0 = const()[name = tensor("v_47_pad_type_0"), val = tensor("custom")]; tensor v_47_pad_0 = const()[name = tensor("v_47_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(739292672)))]; tensor v_47_cast_fp16 = conv(dilations = var_9612, groups = var_8627, pad = v_47_pad_0, pad_type = v_47_pad_type_0, strides = var_9610, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_47_cast_fp16")]; tensor var_9616_begin_0 = const()[name = tensor("op_9616_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9616_end_0 = const()[name = tensor("op_9616_end_0"), val = tensor([2, 64, 1, 576])]; tensor var_9616_end_mask_0 = const()[name = tensor("op_9616_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9616_cast_fp16 = slice_by_index(begin = var_9616_begin_0, end = var_9616_end_0, end_mask = var_9616_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9616_cast_fp16")]; tensor var_9620_begin_0 = const()[name = tensor("op_9620_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_9620_end_0 = const()[name = tensor("op_9620_end_0"), val = tensor([2, 128, 1, 576])]; tensor var_9620_end_mask_0 = const()[name = tensor("op_9620_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9620_cast_fp16 = slice_by_index(begin = var_9620_begin_0, end = var_9620_end_0, end_mask = var_9620_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9620_cast_fp16")]; tensor var_9624_begin_0 = const()[name = tensor("op_9624_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_9624_end_0 = const()[name = tensor("op_9624_end_0"), val = tensor([2, 192, 1, 576])]; tensor var_9624_end_mask_0 = const()[name = tensor("op_9624_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9624_cast_fp16 = slice_by_index(begin = var_9624_begin_0, end = var_9624_end_0, end_mask = var_9624_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9624_cast_fp16")]; tensor var_9628_begin_0 = const()[name = tensor("op_9628_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_9628_end_0 = const()[name = tensor("op_9628_end_0"), val = tensor([2, 256, 1, 576])]; tensor var_9628_end_mask_0 = const()[name = tensor("op_9628_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9628_cast_fp16 = slice_by_index(begin = var_9628_begin_0, end = var_9628_end_0, end_mask = var_9628_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9628_cast_fp16")]; tensor var_9632_begin_0 = const()[name = tensor("op_9632_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_9632_end_0 = const()[name = tensor("op_9632_end_0"), val = tensor([2, 320, 1, 576])]; tensor var_9632_end_mask_0 = const()[name = tensor("op_9632_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9632_cast_fp16 = slice_by_index(begin = var_9632_begin_0, end = var_9632_end_0, end_mask = var_9632_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9632_cast_fp16")]; tensor var_9636_begin_0 = const()[name = tensor("op_9636_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9636_end_0 = const()[name = tensor("op_9636_end_0"), val = tensor([2, 384, 1, 576])]; tensor var_9636_end_mask_0 = const()[name = tensor("op_9636_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9636_cast_fp16 = slice_by_index(begin = var_9636_begin_0, end = var_9636_end_0, end_mask = var_9636_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9636_cast_fp16")]; tensor var_9640_begin_0 = const()[name = tensor("op_9640_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_9640_end_0 = const()[name = tensor("op_9640_end_0"), val = tensor([2, 448, 1, 576])]; tensor var_9640_end_mask_0 = const()[name = tensor("op_9640_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9640_cast_fp16 = slice_by_index(begin = var_9640_begin_0, end = var_9640_end_0, end_mask = var_9640_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9640_cast_fp16")]; tensor var_9644_begin_0 = const()[name = tensor("op_9644_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_9644_end_0 = const()[name = tensor("op_9644_end_0"), val = tensor([2, 512, 1, 576])]; tensor var_9644_end_mask_0 = const()[name = tensor("op_9644_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9644_cast_fp16 = slice_by_index(begin = var_9644_begin_0, end = var_9644_end_0, end_mask = var_9644_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9644_cast_fp16")]; tensor var_9648_begin_0 = const()[name = tensor("op_9648_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_9648_end_0 = const()[name = tensor("op_9648_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_9648_end_mask_0 = const()[name = tensor("op_9648_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9648_cast_fp16 = slice_by_index(begin = var_9648_begin_0, end = var_9648_end_0, end_mask = var_9648_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9648_cast_fp16")]; tensor var_9652_begin_0 = const()[name = tensor("op_9652_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_9652_end_0 = const()[name = tensor("op_9652_end_0"), val = tensor([2, 640, 1, 576])]; tensor var_9652_end_mask_0 = const()[name = tensor("op_9652_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9652_cast_fp16 = slice_by_index(begin = var_9652_begin_0, end = var_9652_end_0, end_mask = var_9652_end_mask_0, x = q_47_cast_fp16)[name = tensor("op_9652_cast_fp16")]; tensor k_95_perm_0 = const()[name = tensor("k_95_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_9659_begin_0 = const()[name = tensor("op_9659_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9659_end_0 = const()[name = tensor("op_9659_end_0"), val = tensor([2, 77, 1, 64])]; tensor var_9659_end_mask_0 = const()[name = tensor("op_9659_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_8 = transpose(perm = k_95_perm_0, x = k_93_cast_fp16)[name = tensor("transpose_8")]; tensor var_9659_cast_fp16 = slice_by_index(begin = var_9659_begin_0, end = var_9659_end_0, end_mask = var_9659_end_mask_0, x = transpose_8)[name = tensor("op_9659_cast_fp16")]; tensor var_9663_begin_0 = const()[name = tensor("op_9663_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_9663_end_0 = const()[name = tensor("op_9663_end_0"), val = tensor([2, 77, 1, 128])]; tensor var_9663_end_mask_0 = const()[name = tensor("op_9663_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9663_cast_fp16 = slice_by_index(begin = var_9663_begin_0, end = var_9663_end_0, end_mask = var_9663_end_mask_0, x = transpose_8)[name = tensor("op_9663_cast_fp16")]; tensor var_9667_begin_0 = const()[name = tensor("op_9667_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_9667_end_0 = const()[name = tensor("op_9667_end_0"), val = tensor([2, 77, 1, 192])]; tensor var_9667_end_mask_0 = const()[name = tensor("op_9667_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9667_cast_fp16 = slice_by_index(begin = var_9667_begin_0, end = var_9667_end_0, end_mask = var_9667_end_mask_0, x = transpose_8)[name = tensor("op_9667_cast_fp16")]; tensor var_9671_begin_0 = const()[name = tensor("op_9671_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_9671_end_0 = const()[name = tensor("op_9671_end_0"), val = tensor([2, 77, 1, 256])]; tensor var_9671_end_mask_0 = const()[name = tensor("op_9671_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9671_cast_fp16 = slice_by_index(begin = var_9671_begin_0, end = var_9671_end_0, end_mask = var_9671_end_mask_0, x = transpose_8)[name = tensor("op_9671_cast_fp16")]; tensor var_9675_begin_0 = const()[name = tensor("op_9675_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_9675_end_0 = const()[name = tensor("op_9675_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_9675_end_mask_0 = const()[name = tensor("op_9675_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9675_cast_fp16 = slice_by_index(begin = var_9675_begin_0, end = var_9675_end_0, end_mask = var_9675_end_mask_0, x = transpose_8)[name = tensor("op_9675_cast_fp16")]; tensor var_9679_begin_0 = const()[name = tensor("op_9679_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_9679_end_0 = const()[name = tensor("op_9679_end_0"), val = tensor([2, 77, 1, 384])]; tensor var_9679_end_mask_0 = const()[name = tensor("op_9679_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9679_cast_fp16 = slice_by_index(begin = var_9679_begin_0, end = var_9679_end_0, end_mask = var_9679_end_mask_0, x = transpose_8)[name = tensor("op_9679_cast_fp16")]; tensor var_9683_begin_0 = const()[name = tensor("op_9683_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_9683_end_0 = const()[name = tensor("op_9683_end_0"), val = tensor([2, 77, 1, 448])]; tensor var_9683_end_mask_0 = const()[name = tensor("op_9683_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9683_cast_fp16 = slice_by_index(begin = var_9683_begin_0, end = var_9683_end_0, end_mask = var_9683_end_mask_0, x = transpose_8)[name = tensor("op_9683_cast_fp16")]; tensor var_9687_begin_0 = const()[name = tensor("op_9687_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_9687_end_0 = const()[name = tensor("op_9687_end_0"), val = tensor([2, 77, 1, 512])]; tensor var_9687_end_mask_0 = const()[name = tensor("op_9687_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9687_cast_fp16 = slice_by_index(begin = var_9687_begin_0, end = var_9687_end_0, end_mask = var_9687_end_mask_0, x = transpose_8)[name = tensor("op_9687_cast_fp16")]; tensor var_9691_begin_0 = const()[name = tensor("op_9691_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9691_end_0 = const()[name = tensor("op_9691_end_0"), val = tensor([2, 77, 1, 576])]; tensor var_9691_end_mask_0 = const()[name = tensor("op_9691_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9691_cast_fp16 = slice_by_index(begin = var_9691_begin_0, end = var_9691_end_0, end_mask = var_9691_end_mask_0, x = transpose_8)[name = tensor("op_9691_cast_fp16")]; tensor var_9695_begin_0 = const()[name = tensor("op_9695_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_9695_end_0 = const()[name = tensor("op_9695_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_9695_end_mask_0 = const()[name = tensor("op_9695_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9695_cast_fp16 = slice_by_index(begin = var_9695_begin_0, end = var_9695_end_0, end_mask = var_9695_end_mask_0, x = transpose_8)[name = tensor("op_9695_cast_fp16")]; tensor var_9697_begin_0 = const()[name = tensor("op_9697_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9697_end_0 = const()[name = tensor("op_9697_end_0"), val = tensor([2, 64, 1, 77])]; tensor var_9697_end_mask_0 = const()[name = tensor("op_9697_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9697_cast_fp16 = slice_by_index(begin = var_9697_begin_0, end = var_9697_end_0, end_mask = var_9697_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9697_cast_fp16")]; tensor var_9701_begin_0 = const()[name = tensor("op_9701_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_9701_end_0 = const()[name = tensor("op_9701_end_0"), val = tensor([2, 128, 1, 77])]; tensor var_9701_end_mask_0 = const()[name = tensor("op_9701_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9701_cast_fp16 = slice_by_index(begin = var_9701_begin_0, end = var_9701_end_0, end_mask = var_9701_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9701_cast_fp16")]; tensor var_9705_begin_0 = const()[name = tensor("op_9705_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_9705_end_0 = const()[name = tensor("op_9705_end_0"), val = tensor([2, 192, 1, 77])]; tensor var_9705_end_mask_0 = const()[name = tensor("op_9705_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9705_cast_fp16 = slice_by_index(begin = var_9705_begin_0, end = var_9705_end_0, end_mask = var_9705_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9705_cast_fp16")]; tensor var_9709_begin_0 = const()[name = tensor("op_9709_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_9709_end_0 = const()[name = tensor("op_9709_end_0"), val = tensor([2, 256, 1, 77])]; tensor var_9709_end_mask_0 = const()[name = tensor("op_9709_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9709_cast_fp16 = slice_by_index(begin = var_9709_begin_0, end = var_9709_end_0, end_mask = var_9709_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9709_cast_fp16")]; tensor var_9713_begin_0 = const()[name = tensor("op_9713_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_9713_end_0 = const()[name = tensor("op_9713_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_9713_end_mask_0 = const()[name = tensor("op_9713_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9713_cast_fp16 = slice_by_index(begin = var_9713_begin_0, end = var_9713_end_0, end_mask = var_9713_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9713_cast_fp16")]; tensor var_9717_begin_0 = const()[name = tensor("op_9717_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9717_end_0 = const()[name = tensor("op_9717_end_0"), val = tensor([2, 384, 1, 77])]; tensor var_9717_end_mask_0 = const()[name = tensor("op_9717_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9717_cast_fp16 = slice_by_index(begin = var_9717_begin_0, end = var_9717_end_0, end_mask = var_9717_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9717_cast_fp16")]; tensor var_9721_begin_0 = const()[name = tensor("op_9721_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_9721_end_0 = const()[name = tensor("op_9721_end_0"), val = tensor([2, 448, 1, 77])]; tensor var_9721_end_mask_0 = const()[name = tensor("op_9721_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9721_cast_fp16 = slice_by_index(begin = var_9721_begin_0, end = var_9721_end_0, end_mask = var_9721_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9721_cast_fp16")]; tensor var_9725_begin_0 = const()[name = tensor("op_9725_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_9725_end_0 = const()[name = tensor("op_9725_end_0"), val = tensor([2, 512, 1, 77])]; tensor var_9725_end_mask_0 = const()[name = tensor("op_9725_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9725_cast_fp16 = slice_by_index(begin = var_9725_begin_0, end = var_9725_end_0, end_mask = var_9725_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9725_cast_fp16")]; tensor var_9729_begin_0 = const()[name = tensor("op_9729_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_9729_end_0 = const()[name = tensor("op_9729_end_0"), val = tensor([2, 576, 1, 77])]; tensor var_9729_end_mask_0 = const()[name = tensor("op_9729_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9729_cast_fp16 = slice_by_index(begin = var_9729_begin_0, end = var_9729_end_0, end_mask = var_9729_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9729_cast_fp16")]; tensor var_9733_begin_0 = const()[name = tensor("op_9733_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_9733_end_0 = const()[name = tensor("op_9733_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_9733_end_mask_0 = const()[name = tensor("op_9733_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9733_cast_fp16 = slice_by_index(begin = var_9733_begin_0, end = var_9733_end_0, end_mask = var_9733_end_mask_0, x = v_47_cast_fp16)[name = tensor("op_9733_cast_fp16")]; tensor var_9737_equation_0 = const()[name = tensor("op_9737_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9737_cast_fp16 = einsum(equation = var_9737_equation_0, values = (var_9659_cast_fp16, var_9616_cast_fp16))[name = tensor("op_9737_cast_fp16")]; tensor var_9738_to_fp16 = const()[name = tensor("op_9738_to_fp16"), val = tensor(0x1p-3)]; tensor aw_661_cast_fp16 = mul(x = var_9737_cast_fp16, y = var_9738_to_fp16)[name = tensor("aw_661_cast_fp16")]; tensor var_9741_equation_0 = const()[name = tensor("op_9741_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9741_cast_fp16 = einsum(equation = var_9741_equation_0, values = (var_9663_cast_fp16, var_9620_cast_fp16))[name = tensor("op_9741_cast_fp16")]; tensor var_9742_to_fp16 = const()[name = tensor("op_9742_to_fp16"), val = tensor(0x1p-3)]; tensor aw_663_cast_fp16 = mul(x = var_9741_cast_fp16, y = var_9742_to_fp16)[name = tensor("aw_663_cast_fp16")]; tensor var_9745_equation_0 = const()[name = tensor("op_9745_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9745_cast_fp16 = einsum(equation = var_9745_equation_0, values = (var_9667_cast_fp16, var_9624_cast_fp16))[name = tensor("op_9745_cast_fp16")]; tensor var_9746_to_fp16 = const()[name = tensor("op_9746_to_fp16"), val = tensor(0x1p-3)]; tensor aw_665_cast_fp16 = mul(x = var_9745_cast_fp16, y = var_9746_to_fp16)[name = tensor("aw_665_cast_fp16")]; tensor var_9749_equation_0 = const()[name = tensor("op_9749_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9749_cast_fp16 = einsum(equation = var_9749_equation_0, values = (var_9671_cast_fp16, var_9628_cast_fp16))[name = tensor("op_9749_cast_fp16")]; tensor var_9750_to_fp16 = const()[name = tensor("op_9750_to_fp16"), val = tensor(0x1p-3)]; tensor aw_667_cast_fp16 = mul(x = var_9749_cast_fp16, y = var_9750_to_fp16)[name = tensor("aw_667_cast_fp16")]; tensor var_9753_equation_0 = const()[name = tensor("op_9753_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9753_cast_fp16 = einsum(equation = var_9753_equation_0, values = (var_9675_cast_fp16, var_9632_cast_fp16))[name = tensor("op_9753_cast_fp16")]; tensor var_9754_to_fp16 = const()[name = tensor("op_9754_to_fp16"), val = tensor(0x1p-3)]; tensor aw_669_cast_fp16 = mul(x = var_9753_cast_fp16, y = var_9754_to_fp16)[name = tensor("aw_669_cast_fp16")]; tensor var_9757_equation_0 = const()[name = tensor("op_9757_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9757_cast_fp16 = einsum(equation = var_9757_equation_0, values = (var_9679_cast_fp16, var_9636_cast_fp16))[name = tensor("op_9757_cast_fp16")]; tensor var_9758_to_fp16 = const()[name = tensor("op_9758_to_fp16"), val = tensor(0x1p-3)]; tensor aw_671_cast_fp16 = mul(x = var_9757_cast_fp16, y = var_9758_to_fp16)[name = tensor("aw_671_cast_fp16")]; tensor var_9761_equation_0 = const()[name = tensor("op_9761_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9761_cast_fp16 = einsum(equation = var_9761_equation_0, values = (var_9683_cast_fp16, var_9640_cast_fp16))[name = tensor("op_9761_cast_fp16")]; tensor var_9762_to_fp16 = const()[name = tensor("op_9762_to_fp16"), val = tensor(0x1p-3)]; tensor aw_673_cast_fp16 = mul(x = var_9761_cast_fp16, y = var_9762_to_fp16)[name = tensor("aw_673_cast_fp16")]; tensor var_9765_equation_0 = const()[name = tensor("op_9765_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9765_cast_fp16 = einsum(equation = var_9765_equation_0, values = (var_9687_cast_fp16, var_9644_cast_fp16))[name = tensor("op_9765_cast_fp16")]; tensor var_9766_to_fp16 = const()[name = tensor("op_9766_to_fp16"), val = tensor(0x1p-3)]; tensor aw_675_cast_fp16 = mul(x = var_9765_cast_fp16, y = var_9766_to_fp16)[name = tensor("aw_675_cast_fp16")]; tensor var_9769_equation_0 = const()[name = tensor("op_9769_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9769_cast_fp16 = einsum(equation = var_9769_equation_0, values = (var_9691_cast_fp16, var_9648_cast_fp16))[name = tensor("op_9769_cast_fp16")]; tensor var_9770_to_fp16 = const()[name = tensor("op_9770_to_fp16"), val = tensor(0x1p-3)]; tensor aw_677_cast_fp16 = mul(x = var_9769_cast_fp16, y = var_9770_to_fp16)[name = tensor("aw_677_cast_fp16")]; tensor var_9773_equation_0 = const()[name = tensor("op_9773_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9773_cast_fp16 = einsum(equation = var_9773_equation_0, values = (var_9695_cast_fp16, var_9652_cast_fp16))[name = tensor("op_9773_cast_fp16")]; tensor var_9774_to_fp16 = const()[name = tensor("op_9774_to_fp16"), val = tensor(0x1p-3)]; tensor aw_679_cast_fp16 = mul(x = var_9773_cast_fp16, y = var_9774_to_fp16)[name = tensor("aw_679_cast_fp16")]; tensor var_9776_cast_fp16 = softmax(axis = var_8627, x = aw_661_cast_fp16)[name = tensor("op_9776_cast_fp16")]; tensor var_9777_cast_fp16 = softmax(axis = var_8627, x = aw_663_cast_fp16)[name = tensor("op_9777_cast_fp16")]; tensor var_9778_cast_fp16 = softmax(axis = var_8627, x = aw_665_cast_fp16)[name = tensor("op_9778_cast_fp16")]; tensor var_9779_cast_fp16 = softmax(axis = var_8627, x = aw_667_cast_fp16)[name = tensor("op_9779_cast_fp16")]; tensor var_9780_cast_fp16 = softmax(axis = var_8627, x = aw_669_cast_fp16)[name = tensor("op_9780_cast_fp16")]; tensor var_9781_cast_fp16 = softmax(axis = var_8627, x = aw_671_cast_fp16)[name = tensor("op_9781_cast_fp16")]; tensor var_9782_cast_fp16 = softmax(axis = var_8627, x = aw_673_cast_fp16)[name = tensor("op_9782_cast_fp16")]; tensor var_9783_cast_fp16 = softmax(axis = var_8627, x = aw_675_cast_fp16)[name = tensor("op_9783_cast_fp16")]; tensor var_9784_cast_fp16 = softmax(axis = var_8627, x = aw_677_cast_fp16)[name = tensor("op_9784_cast_fp16")]; tensor var_9785_cast_fp16 = softmax(axis = var_8627, x = aw_679_cast_fp16)[name = tensor("op_9785_cast_fp16")]; tensor var_9787_equation_0 = const()[name = tensor("op_9787_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9787_cast_fp16 = einsum(equation = var_9787_equation_0, values = (var_9697_cast_fp16, var_9776_cast_fp16))[name = tensor("op_9787_cast_fp16")]; tensor var_9789_equation_0 = const()[name = tensor("op_9789_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9789_cast_fp16 = einsum(equation = var_9789_equation_0, values = (var_9701_cast_fp16, var_9777_cast_fp16))[name = tensor("op_9789_cast_fp16")]; tensor var_9791_equation_0 = const()[name = tensor("op_9791_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9791_cast_fp16 = einsum(equation = var_9791_equation_0, values = (var_9705_cast_fp16, var_9778_cast_fp16))[name = tensor("op_9791_cast_fp16")]; tensor var_9793_equation_0 = const()[name = tensor("op_9793_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9793_cast_fp16 = einsum(equation = var_9793_equation_0, values = (var_9709_cast_fp16, var_9779_cast_fp16))[name = tensor("op_9793_cast_fp16")]; tensor var_9795_equation_0 = const()[name = tensor("op_9795_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9795_cast_fp16 = einsum(equation = var_9795_equation_0, values = (var_9713_cast_fp16, var_9780_cast_fp16))[name = tensor("op_9795_cast_fp16")]; tensor var_9797_equation_0 = const()[name = tensor("op_9797_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9797_cast_fp16 = einsum(equation = var_9797_equation_0, values = (var_9717_cast_fp16, var_9781_cast_fp16))[name = tensor("op_9797_cast_fp16")]; tensor var_9799_equation_0 = const()[name = tensor("op_9799_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9799_cast_fp16 = einsum(equation = var_9799_equation_0, values = (var_9721_cast_fp16, var_9782_cast_fp16))[name = tensor("op_9799_cast_fp16")]; tensor var_9801_equation_0 = const()[name = tensor("op_9801_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9801_cast_fp16 = einsum(equation = var_9801_equation_0, values = (var_9725_cast_fp16, var_9783_cast_fp16))[name = tensor("op_9801_cast_fp16")]; tensor var_9803_equation_0 = const()[name = tensor("op_9803_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9803_cast_fp16 = einsum(equation = var_9803_equation_0, values = (var_9729_cast_fp16, var_9784_cast_fp16))[name = tensor("op_9803_cast_fp16")]; tensor var_9805_equation_0 = const()[name = tensor("op_9805_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9805_cast_fp16 = einsum(equation = var_9805_equation_0, values = (var_9733_cast_fp16, var_9785_cast_fp16))[name = tensor("op_9805_cast_fp16")]; tensor input_411_interleave_0 = const()[name = tensor("input_411_interleave_0"), val = tensor(false)]; tensor input_411_cast_fp16 = concat(axis = var_8627, interleave = input_411_interleave_0, values = (var_9787_cast_fp16, var_9789_cast_fp16, var_9791_cast_fp16, var_9793_cast_fp16, var_9795_cast_fp16, var_9797_cast_fp16, var_9799_cast_fp16, var_9801_cast_fp16, var_9803_cast_fp16, var_9805_cast_fp16))[name = tensor("input_411_cast_fp16")]; tensor var_9811 = const()[name = tensor("op_9811"), val = tensor([1, 1])]; tensor var_9813 = const()[name = tensor("op_9813"), val = tensor([1, 1])]; tensor var_9815_pad_type_0 = const()[name = tensor("op_9815_pad_type_0"), val = tensor("custom")]; tensor var_9815_pad_0 = const()[name = tensor("op_9815_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(740603456)))]; tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(741422720)))]; tensor var_9815_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_9813, groups = var_8627, pad = var_9815_pad_0, pad_type = var_9815_pad_type_0, strides = var_9811, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_411_cast_fp16)[name = tensor("op_9815_cast_fp16")]; tensor inputs_71_cast_fp16 = add(x = var_9815_cast_fp16, y = inputs_69_cast_fp16)[name = tensor("inputs_71_cast_fp16")]; tensor var_9819 = const()[name = tensor("op_9819"), val = tensor([1])]; tensor channels_mean_71_cast_fp16 = reduce_mean(axes = var_9819, keep_dims = var_8622, x = inputs_71_cast_fp16)[name = tensor("channels_mean_71_cast_fp16")]; tensor zero_mean_71_cast_fp16 = sub(x = inputs_71_cast_fp16, y = channels_mean_71_cast_fp16)[name = tensor("zero_mean_71_cast_fp16")]; tensor zero_mean_sq_71_cast_fp16 = mul(x = zero_mean_71_cast_fp16, y = zero_mean_71_cast_fp16)[name = tensor("zero_mean_sq_71_cast_fp16")]; tensor var_9823 = const()[name = tensor("op_9823"), val = tensor([1])]; tensor var_9824_cast_fp16 = reduce_mean(axes = var_9823, keep_dims = var_8622, x = zero_mean_sq_71_cast_fp16)[name = tensor("op_9824_cast_fp16")]; tensor var_9825_to_fp16 = const()[name = tensor("op_9825_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_9826_cast_fp16 = add(x = var_9824_cast_fp16, y = var_9825_to_fp16)[name = tensor("op_9826_cast_fp16")]; tensor denom_71_epsilon_0_to_fp16 = const()[name = tensor("denom_71_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_71_cast_fp16 = rsqrt(epsilon = denom_71_epsilon_0_to_fp16, x = var_9826_cast_fp16)[name = tensor("denom_71_cast_fp16")]; tensor out_71_cast_fp16 = mul(x = zero_mean_71_cast_fp16, y = denom_71_cast_fp16)[name = tensor("out_71_cast_fp16")]; tensor var_9830_to_fp16 = const()[name = tensor("op_9830_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(741424064)))]; tensor var_9831_cast_fp16 = add(x = out_71_cast_fp16, y = var_9830_to_fp16)[name = tensor("op_9831_cast_fp16")]; tensor var_9833_to_fp16 = const()[name = tensor("op_9833_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(741425408)))]; tensor input_413_cast_fp16 = mul(x = var_9831_cast_fp16, y = var_9833_to_fp16)[name = tensor("input_413_cast_fp16")]; tensor var_9841 = const()[name = tensor("op_9841"), val = tensor([1, 1])]; tensor var_9843 = const()[name = tensor("op_9843"), val = tensor([1, 1])]; tensor var_9845_pad_type_0 = const()[name = tensor("op_9845_pad_type_0"), val = tensor("custom")]; tensor var_9845_pad_0 = const()[name = tensor("op_9845_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(741426752)))]; tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(747980416)))]; tensor var_9845_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_9843, groups = var_8627, pad = var_9845_pad_0, pad_type = var_9845_pad_type_0, strides = var_9841, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_413_cast_fp16)[name = tensor("op_9845_cast_fp16")]; tensor var_9846_split_sizes_0 = const()[name = tensor("op_9846_split_sizes_0"), val = tensor([2560, 2560])]; tensor var_9846_axis_0 = const()[name = tensor("op_9846_axis_0"), val = tensor(1)]; tensor var_9846_cast_fp16_0, tensor var_9846_cast_fp16_1 = split(axis = var_9846_axis_0, split_sizes = var_9846_split_sizes_0, x = var_9845_cast_fp16)[name = tensor("op_9846_cast_fp16")]; tensor var_9848_mode_0 = const()[name = tensor("op_9848_mode_0"), val = tensor("EXACT")]; tensor var_9848_cast_fp16 = gelu(mode = var_9848_mode_0, x = var_9846_cast_fp16_1)[name = tensor("op_9848_cast_fp16")]; tensor input_415_cast_fp16 = mul(x = var_9846_cast_fp16_0, y = var_9848_cast_fp16)[name = tensor("input_415_cast_fp16")]; tensor var_9852 = const()[name = tensor("op_9852"), val = tensor([1, 1])]; tensor var_9854 = const()[name = tensor("op_9854"), val = tensor([1, 1])]; tensor var_9856_pad_type_0 = const()[name = tensor("op_9856_pad_type_0"), val = tensor("custom")]; tensor var_9856_pad_0 = const()[name = tensor("op_9856_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(747990720)))]; tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(751267584)))]; tensor var_9856_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_9854, groups = var_8627, pad = var_9856_pad_0, pad_type = var_9856_pad_type_0, strides = var_9852, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_415_cast_fp16)[name = tensor("op_9856_cast_fp16")]; tensor hidden_states_253_cast_fp16 = add(x = var_9856_cast_fp16, y = inputs_71_cast_fp16)[name = tensor("hidden_states_253_cast_fp16")]; tensor var_9858 = const()[name = tensor("op_9858"), val = tensor([2, 640, 24, 24])]; tensor input_417_cast_fp16 = reshape(shape = var_9858, x = hidden_states_253_cast_fp16)[name = tensor("input_417_cast_fp16")]; tensor var_9862 = const()[name = tensor("op_9862"), val = tensor([1, 1])]; tensor var_9864 = const()[name = tensor("op_9864"), val = tensor([1, 1])]; tensor hidden_states_255_pad_type_0 = const()[name = tensor("hidden_states_255_pad_type_0"), val = tensor("custom")]; tensor hidden_states_255_pad_0 = const()[name = tensor("hidden_states_255_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(751268928)))]; tensor up_blocks_2_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(752088192)))]; tensor hidden_states_255_cast_fp16 = conv(bias = up_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_9864, groups = var_8627, pad = hidden_states_255_pad_0, pad_type = hidden_states_255_pad_type_0, strides = var_9862, weight = up_blocks_2_attentions_1_proj_out_weight_to_fp16, x = input_417_cast_fp16)[name = tensor("hidden_states_255_cast_fp16")]; tensor hidden_states_257_cast_fp16 = add(x = hidden_states_255_cast_fp16, y = hidden_states_243_cast_fp16)[name = tensor("hidden_states_257_cast_fp16")]; tensor input_419_interleave_0 = const()[name = tensor("input_419_interleave_0"), val = tensor(false)]; tensor cast_9 = cast(dtype = cast_7_dtype_0, x = input_63_cast_fp16)[name = tensor("cast_9")]; tensor input_419_cast_fp16 = concat(axis = var_8627, interleave = input_419_interleave_0, values = (hidden_states_257_cast_fp16, cast_9))[name = tensor("input_419_cast_fp16")]; tensor reshape_192_shape_0 = const()[name = tensor("reshape_192_shape_0"), val = tensor([2, 32, 30, 24, 24])]; tensor reshape_192_cast_fp16 = reshape(shape = reshape_192_shape_0, x = input_419_cast_fp16)[name = tensor("reshape_192_cast_fp16")]; tensor reduce_mean_144_axes_0 = const()[name = tensor("reduce_mean_144_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_144_keep_dims_0 = const()[name = tensor("reduce_mean_144_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_144_cast_fp16 = reduce_mean(axes = reduce_mean_144_axes_0, keep_dims = reduce_mean_144_keep_dims_0, x = reshape_192_cast_fp16)[name = tensor("reduce_mean_144_cast_fp16")]; tensor sub_96_cast_fp16 = sub(x = reshape_192_cast_fp16, y = reduce_mean_144_cast_fp16)[name = tensor("sub_96_cast_fp16")]; tensor square_48_cast_fp16 = square(x = sub_96_cast_fp16)[name = tensor("square_48_cast_fp16")]; tensor reduce_mean_146_axes_0 = const()[name = tensor("reduce_mean_146_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_146_keep_dims_0 = const()[name = tensor("reduce_mean_146_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_146_cast_fp16 = reduce_mean(axes = reduce_mean_146_axes_0, keep_dims = reduce_mean_146_keep_dims_0, x = square_48_cast_fp16)[name = tensor("reduce_mean_146_cast_fp16")]; tensor add_96_y_0_to_fp16 = const()[name = tensor("add_96_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_96_cast_fp16 = add(x = reduce_mean_146_cast_fp16, y = add_96_y_0_to_fp16)[name = tensor("add_96_cast_fp16")]; tensor sqrt_48_cast_fp16 = sqrt(x = add_96_cast_fp16)[name = tensor("sqrt_48_cast_fp16")]; tensor real_div_48_cast_fp16 = real_div(x = sub_96_cast_fp16, y = sqrt_48_cast_fp16)[name = tensor("real_div_48_cast_fp16")]; tensor reshape_193_shape_0 = const()[name = tensor("reshape_193_shape_0"), val = tensor([2, 960, 24, 24])]; tensor reshape_193_cast_fp16 = reshape(shape = reshape_193_shape_0, x = real_div_48_cast_fp16)[name = tensor("reshape_193_cast_fp16")]; tensor add_97_mean_0_to_fp16 = const()[name = tensor("add_97_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(752089536)))]; tensor add_97_variance_0_to_fp16 = const()[name = tensor("add_97_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(752091520)))]; tensor add_97_gamma_0_to_fp16 = const()[name = tensor("add_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(752093504)))]; tensor add_97_beta_0_to_fp16 = const()[name = tensor("add_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(752095488)))]; tensor add_97_epsilon_0_to_fp16 = const()[name = tensor("add_97_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_97_cast_fp16 = batch_norm(beta = add_97_beta_0_to_fp16, epsilon = add_97_epsilon_0_to_fp16, gamma = add_97_gamma_0_to_fp16, mean = add_97_mean_0_to_fp16, variance = add_97_variance_0_to_fp16, x = reshape_193_cast_fp16)[name = tensor("add_97_cast_fp16")]; tensor input_423_cast_fp16 = silu(x = add_97_cast_fp16)[name = tensor("input_423_cast_fp16")]; tensor var_9882 = const()[name = tensor("op_9882"), val = tensor([1, 1])]; tensor var_9884 = const()[name = tensor("op_9884"), val = tensor([1, 1])]; tensor hidden_states_259_pad_type_0 = const()[name = tensor("hidden_states_259_pad_type_0"), val = tensor("custom")]; tensor hidden_states_259_pad_0 = const()[name = tensor("hidden_states_259_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_2_resnets_2_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(752097472)))]; tensor up_blocks_2_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(763156736)))]; tensor hidden_states_259_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv1_bias_to_fp16, dilations = var_9884, groups = var_8627, pad = hidden_states_259_pad_0, pad_type = hidden_states_259_pad_type_0, strides = var_9882, weight = up_blocks_2_resnets_2_conv1_weight_to_fp16, x = input_423_cast_fp16)[name = tensor("hidden_states_259_cast_fp16")]; tensor var_9890 = const()[name = tensor("op_9890"), val = tensor([1, 1])]; tensor var_9892 = const()[name = tensor("op_9892"), val = tensor([1, 1])]; tensor temb_37_pad_type_0 = const()[name = tensor("temb_37_pad_type_0"), val = tensor("custom")]; tensor temb_37_pad_0 = const()[name = tensor("temb_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(763158080)))]; tensor up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(764796544)))]; tensor temb_37_cast_fp16 = conv(bias = up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_9892, groups = var_8627, pad = temb_37_pad_0, pad_type = temb_37_pad_type_0, strides = var_9890, weight = up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16, x = cast_12)[name = tensor("temb_37_cast_fp16")]; tensor input_427_cast_fp16 = add(x = hidden_states_259_cast_fp16, y = temb_37_cast_fp16)[name = tensor("input_427_cast_fp16")]; tensor reshape_196_shape_0 = const()[name = tensor("reshape_196_shape_0"), val = tensor([2, 32, 20, 24, 24])]; tensor reshape_196_cast_fp16 = reshape(shape = reshape_196_shape_0, x = input_427_cast_fp16)[name = tensor("reshape_196_cast_fp16")]; tensor reduce_mean_147_axes_0 = const()[name = tensor("reduce_mean_147_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_147_keep_dims_0 = const()[name = tensor("reduce_mean_147_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_147_cast_fp16 = reduce_mean(axes = reduce_mean_147_axes_0, keep_dims = reduce_mean_147_keep_dims_0, x = reshape_196_cast_fp16)[name = tensor("reduce_mean_147_cast_fp16")]; tensor sub_98_cast_fp16 = sub(x = reshape_196_cast_fp16, y = reduce_mean_147_cast_fp16)[name = tensor("sub_98_cast_fp16")]; tensor square_49_cast_fp16 = square(x = sub_98_cast_fp16)[name = tensor("square_49_cast_fp16")]; tensor reduce_mean_149_axes_0 = const()[name = tensor("reduce_mean_149_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_149_keep_dims_0 = const()[name = tensor("reduce_mean_149_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_149_cast_fp16 = reduce_mean(axes = reduce_mean_149_axes_0, keep_dims = reduce_mean_149_keep_dims_0, x = square_49_cast_fp16)[name = tensor("reduce_mean_149_cast_fp16")]; tensor add_98_y_0_to_fp16 = const()[name = tensor("add_98_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_98_cast_fp16 = add(x = reduce_mean_149_cast_fp16, y = add_98_y_0_to_fp16)[name = tensor("add_98_cast_fp16")]; tensor sqrt_49_cast_fp16 = sqrt(x = add_98_cast_fp16)[name = tensor("sqrt_49_cast_fp16")]; tensor real_div_49_cast_fp16 = real_div(x = sub_98_cast_fp16, y = sqrt_49_cast_fp16)[name = tensor("real_div_49_cast_fp16")]; tensor reshape_197_shape_0 = const()[name = tensor("reshape_197_shape_0"), val = tensor([2, 640, 24, 24])]; tensor reshape_197_cast_fp16 = reshape(shape = reshape_197_shape_0, x = real_div_49_cast_fp16)[name = tensor("reshape_197_cast_fp16")]; tensor add_99_gamma_0_to_fp16 = const()[name = tensor("add_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(764797888)))]; tensor add_99_beta_0_to_fp16 = const()[name = tensor("add_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(764799232)))]; tensor add_99_epsilon_0_to_fp16 = const()[name = tensor("add_99_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_99_cast_fp16 = batch_norm(beta = add_99_beta_0_to_fp16, epsilon = add_99_epsilon_0_to_fp16, gamma = add_99_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_197_cast_fp16)[name = tensor("add_99_cast_fp16")]; tensor input_431_cast_fp16 = silu(x = add_99_cast_fp16)[name = tensor("input_431_cast_fp16")]; tensor var_9902 = const()[name = tensor("op_9902"), val = tensor([1, 1])]; tensor var_9904 = const()[name = tensor("op_9904"), val = tensor([1, 1])]; tensor hidden_states_261_pad_type_0 = const()[name = tensor("hidden_states_261_pad_type_0"), val = tensor("custom")]; tensor hidden_states_261_pad_0 = const()[name = tensor("hidden_states_261_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_2_resnets_2_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(764800576)))]; tensor up_blocks_2_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(772173440)))]; tensor hidden_states_261_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv2_bias_to_fp16, dilations = var_9904, groups = var_8627, pad = hidden_states_261_pad_0, pad_type = hidden_states_261_pad_type_0, strides = var_9902, weight = up_blocks_2_resnets_2_conv2_weight_to_fp16, x = input_431_cast_fp16)[name = tensor("hidden_states_261_cast_fp16")]; tensor var_9909 = const()[name = tensor("op_9909"), val = tensor([1, 1])]; tensor var_9911 = const()[name = tensor("op_9911"), val = tensor([1, 1])]; tensor x_21_pad_type_0 = const()[name = tensor("x_21_pad_type_0"), val = tensor("custom")]; tensor x_21_pad_0 = const()[name = tensor("x_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(772174784)))]; tensor up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(773403648)))]; tensor x_21_cast_fp16 = conv(bias = up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_9911, groups = var_8627, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = var_9909, weight = up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16, x = input_419_cast_fp16)[name = tensor("x_21_cast_fp16")]; tensor hidden_states_263_cast_fp16 = add(x = x_21_cast_fp16, y = hidden_states_261_cast_fp16)[name = tensor("hidden_states_263_cast_fp16")]; tensor reshape_200_shape_0 = const()[name = tensor("reshape_200_shape_0"), val = tensor([2, 32, 20, 24, 24])]; tensor reshape_200_cast_fp16 = reshape(shape = reshape_200_shape_0, x = hidden_states_263_cast_fp16)[name = tensor("reshape_200_cast_fp16")]; tensor reduce_mean_150_axes_0 = const()[name = tensor("reduce_mean_150_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_150_keep_dims_0 = const()[name = tensor("reduce_mean_150_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_150_cast_fp16 = reduce_mean(axes = reduce_mean_150_axes_0, keep_dims = reduce_mean_150_keep_dims_0, x = reshape_200_cast_fp16)[name = tensor("reduce_mean_150_cast_fp16")]; tensor sub_100_cast_fp16 = sub(x = reshape_200_cast_fp16, y = reduce_mean_150_cast_fp16)[name = tensor("sub_100_cast_fp16")]; tensor square_50_cast_fp16 = square(x = sub_100_cast_fp16)[name = tensor("square_50_cast_fp16")]; tensor reduce_mean_152_axes_0 = const()[name = tensor("reduce_mean_152_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_152_keep_dims_0 = const()[name = tensor("reduce_mean_152_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_152_cast_fp16 = reduce_mean(axes = reduce_mean_152_axes_0, keep_dims = reduce_mean_152_keep_dims_0, x = square_50_cast_fp16)[name = tensor("reduce_mean_152_cast_fp16")]; tensor add_100_y_0_to_fp16 = const()[name = tensor("add_100_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_100_cast_fp16 = add(x = reduce_mean_152_cast_fp16, y = add_100_y_0_to_fp16)[name = tensor("add_100_cast_fp16")]; tensor sqrt_50_cast_fp16 = sqrt(x = add_100_cast_fp16)[name = tensor("sqrt_50_cast_fp16")]; tensor real_div_50_cast_fp16 = real_div(x = sub_100_cast_fp16, y = sqrt_50_cast_fp16)[name = tensor("real_div_50_cast_fp16")]; tensor reshape_201_shape_0 = const()[name = tensor("reshape_201_shape_0"), val = tensor([2, 640, 24, 24])]; tensor reshape_201_cast_fp16 = reshape(shape = reshape_201_shape_0, x = real_div_50_cast_fp16)[name = tensor("reshape_201_cast_fp16")]; tensor add_101_gamma_0_to_fp16 = const()[name = tensor("add_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(773404992)))]; tensor add_101_beta_0_to_fp16 = const()[name = tensor("add_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(773406336)))]; tensor add_101_epsilon_0_to_fp16 = const()[name = tensor("add_101_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_101_cast_fp16 = batch_norm(beta = add_101_beta_0_to_fp16, epsilon = add_101_epsilon_0_to_fp16, gamma = add_101_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_201_cast_fp16)[name = tensor("add_101_cast_fp16")]; tensor var_9931 = const()[name = tensor("op_9931"), val = tensor([1, 1])]; tensor var_9933 = const()[name = tensor("op_9933"), val = tensor([1, 1])]; tensor hidden_states_265_pad_type_0 = const()[name = tensor("hidden_states_265_pad_type_0"), val = tensor("custom")]; tensor hidden_states_265_pad_0 = const()[name = tensor("hidden_states_265_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(773407680)))]; tensor up_blocks_2_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(774226944)))]; tensor hidden_states_265_cast_fp16 = conv(bias = up_blocks_2_attentions_2_proj_in_bias_to_fp16, dilations = var_9933, groups = var_8627, pad = hidden_states_265_pad_0, pad_type = hidden_states_265_pad_type_0, strides = var_9931, weight = up_blocks_2_attentions_2_proj_in_weight_to_fp16, x = add_101_cast_fp16)[name = tensor("hidden_states_265_cast_fp16")]; tensor var_9938 = const()[name = tensor("op_9938"), val = tensor([2, 640, 1, 576])]; tensor inputs_73_cast_fp16 = reshape(shape = var_9938, x = hidden_states_265_cast_fp16)[name = tensor("inputs_73_cast_fp16")]; tensor var_9948 = const()[name = tensor("op_9948"), val = tensor([1])]; tensor channels_mean_73_cast_fp16 = reduce_mean(axes = var_9948, keep_dims = var_8622, x = inputs_73_cast_fp16)[name = tensor("channels_mean_73_cast_fp16")]; tensor zero_mean_73_cast_fp16 = sub(x = inputs_73_cast_fp16, y = channels_mean_73_cast_fp16)[name = tensor("zero_mean_73_cast_fp16")]; tensor zero_mean_sq_73_cast_fp16 = mul(x = zero_mean_73_cast_fp16, y = zero_mean_73_cast_fp16)[name = tensor("zero_mean_sq_73_cast_fp16")]; tensor var_9952 = const()[name = tensor("op_9952"), val = tensor([1])]; tensor var_9953_cast_fp16 = reduce_mean(axes = var_9952, keep_dims = var_8622, x = zero_mean_sq_73_cast_fp16)[name = tensor("op_9953_cast_fp16")]; tensor var_9954_to_fp16 = const()[name = tensor("op_9954_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_9955_cast_fp16 = add(x = var_9953_cast_fp16, y = var_9954_to_fp16)[name = tensor("op_9955_cast_fp16")]; tensor denom_73_epsilon_0_to_fp16 = const()[name = tensor("denom_73_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_73_cast_fp16 = rsqrt(epsilon = denom_73_epsilon_0_to_fp16, x = var_9955_cast_fp16)[name = tensor("denom_73_cast_fp16")]; tensor out_73_cast_fp16 = mul(x = zero_mean_73_cast_fp16, y = denom_73_cast_fp16)[name = tensor("out_73_cast_fp16")]; tensor var_9959_to_fp16 = const()[name = tensor("op_9959_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(774228288)))]; tensor var_9960_cast_fp16 = add(x = out_73_cast_fp16, y = var_9959_to_fp16)[name = tensor("op_9960_cast_fp16")]; tensor var_9962_to_fp16 = const()[name = tensor("op_9962_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(774229632)))]; tensor hidden_states_267_cast_fp16 = mul(x = var_9960_cast_fp16, y = var_9962_to_fp16)[name = tensor("hidden_states_267_cast_fp16")]; tensor var_9969 = const()[name = tensor("op_9969"), val = tensor([1, 1])]; tensor var_9971 = const()[name = tensor("op_9971"), val = tensor([1, 1])]; tensor q_49_pad_type_0 = const()[name = tensor("q_49_pad_type_0"), val = tensor("custom")]; tensor q_49_pad_0 = const()[name = tensor("q_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(774230976)))]; tensor q_49_cast_fp16 = conv(dilations = var_9971, groups = var_8627, pad = q_49_pad_0, pad_type = q_49_pad_type_0, strides = var_9969, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_267_cast_fp16)[name = tensor("q_49_cast_fp16")]; tensor var_9975 = const()[name = tensor("op_9975"), val = tensor([1, 1])]; tensor var_9977 = const()[name = tensor("op_9977"), val = tensor([1, 1])]; tensor k_97_pad_type_0 = const()[name = tensor("k_97_pad_type_0"), val = tensor("custom")]; tensor k_97_pad_0 = const()[name = tensor("k_97_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775050240)))]; tensor k_97_cast_fp16 = conv(dilations = var_9977, groups = var_8627, pad = k_97_pad_0, pad_type = k_97_pad_type_0, strides = var_9975, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_267_cast_fp16)[name = tensor("k_97_cast_fp16")]; tensor var_9981 = const()[name = tensor("op_9981"), val = tensor([1, 1])]; tensor var_9983 = const()[name = tensor("op_9983"), val = tensor([1, 1])]; tensor v_49_pad_type_0 = const()[name = tensor("v_49_pad_type_0"), val = tensor("custom")]; tensor v_49_pad_0 = const()[name = tensor("v_49_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775869504)))]; tensor v_49_cast_fp16 = conv(dilations = var_9983, groups = var_8627, pad = v_49_pad_0, pad_type = v_49_pad_type_0, strides = var_9981, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_267_cast_fp16)[name = tensor("v_49_cast_fp16")]; tensor var_9987_begin_0 = const()[name = tensor("op_9987_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9987_end_0 = const()[name = tensor("op_9987_end_0"), val = tensor([2, 64, 1, 576])]; tensor var_9987_end_mask_0 = const()[name = tensor("op_9987_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9987_cast_fp16 = slice_by_index(begin = var_9987_begin_0, end = var_9987_end_0, end_mask = var_9987_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9987_cast_fp16")]; tensor var_9991_begin_0 = const()[name = tensor("op_9991_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_9991_end_0 = const()[name = tensor("op_9991_end_0"), val = tensor([2, 128, 1, 576])]; tensor var_9991_end_mask_0 = const()[name = tensor("op_9991_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9991_cast_fp16 = slice_by_index(begin = var_9991_begin_0, end = var_9991_end_0, end_mask = var_9991_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9991_cast_fp16")]; tensor var_9995_begin_0 = const()[name = tensor("op_9995_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_9995_end_0 = const()[name = tensor("op_9995_end_0"), val = tensor([2, 192, 1, 576])]; tensor var_9995_end_mask_0 = const()[name = tensor("op_9995_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9995_cast_fp16 = slice_by_index(begin = var_9995_begin_0, end = var_9995_end_0, end_mask = var_9995_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9995_cast_fp16")]; tensor var_9999_begin_0 = const()[name = tensor("op_9999_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_9999_end_0 = const()[name = tensor("op_9999_end_0"), val = tensor([2, 256, 1, 576])]; tensor var_9999_end_mask_0 = const()[name = tensor("op_9999_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9999_cast_fp16 = slice_by_index(begin = var_9999_begin_0, end = var_9999_end_0, end_mask = var_9999_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_9999_cast_fp16")]; tensor var_10003_begin_0 = const()[name = tensor("op_10003_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10003_end_0 = const()[name = tensor("op_10003_end_0"), val = tensor([2, 320, 1, 576])]; tensor var_10003_end_mask_0 = const()[name = tensor("op_10003_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10003_cast_fp16 = slice_by_index(begin = var_10003_begin_0, end = var_10003_end_0, end_mask = var_10003_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10003_cast_fp16")]; tensor var_10007_begin_0 = const()[name = tensor("op_10007_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_10007_end_0 = const()[name = tensor("op_10007_end_0"), val = tensor([2, 384, 1, 576])]; tensor var_10007_end_mask_0 = const()[name = tensor("op_10007_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10007_cast_fp16 = slice_by_index(begin = var_10007_begin_0, end = var_10007_end_0, end_mask = var_10007_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10007_cast_fp16")]; tensor var_10011_begin_0 = const()[name = tensor("op_10011_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_10011_end_0 = const()[name = tensor("op_10011_end_0"), val = tensor([2, 448, 1, 576])]; tensor var_10011_end_mask_0 = const()[name = tensor("op_10011_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10011_cast_fp16 = slice_by_index(begin = var_10011_begin_0, end = var_10011_end_0, end_mask = var_10011_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10011_cast_fp16")]; tensor var_10015_begin_0 = const()[name = tensor("op_10015_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_10015_end_0 = const()[name = tensor("op_10015_end_0"), val = tensor([2, 512, 1, 576])]; tensor var_10015_end_mask_0 = const()[name = tensor("op_10015_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10015_cast_fp16 = slice_by_index(begin = var_10015_begin_0, end = var_10015_end_0, end_mask = var_10015_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10015_cast_fp16")]; tensor var_10019_begin_0 = const()[name = tensor("op_10019_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_10019_end_0 = const()[name = tensor("op_10019_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_10019_end_mask_0 = const()[name = tensor("op_10019_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10019_cast_fp16 = slice_by_index(begin = var_10019_begin_0, end = var_10019_end_0, end_mask = var_10019_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10019_cast_fp16")]; tensor var_10023_begin_0 = const()[name = tensor("op_10023_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_10023_end_0 = const()[name = tensor("op_10023_end_0"), val = tensor([2, 640, 1, 576])]; tensor var_10023_end_mask_0 = const()[name = tensor("op_10023_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10023_cast_fp16 = slice_by_index(begin = var_10023_begin_0, end = var_10023_end_0, end_mask = var_10023_end_mask_0, x = q_49_cast_fp16)[name = tensor("op_10023_cast_fp16")]; tensor k_99_perm_0 = const()[name = tensor("k_99_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_10030_begin_0 = const()[name = tensor("op_10030_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10030_end_0 = const()[name = tensor("op_10030_end_0"), val = tensor([2, 576, 1, 64])]; tensor var_10030_end_mask_0 = const()[name = tensor("op_10030_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_7 = transpose(perm = k_99_perm_0, x = k_97_cast_fp16)[name = tensor("transpose_7")]; tensor var_10030_cast_fp16 = slice_by_index(begin = var_10030_begin_0, end = var_10030_end_0, end_mask = var_10030_end_mask_0, x = transpose_7)[name = tensor("op_10030_cast_fp16")]; tensor var_10034_begin_0 = const()[name = tensor("op_10034_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_10034_end_0 = const()[name = tensor("op_10034_end_0"), val = tensor([2, 576, 1, 128])]; tensor var_10034_end_mask_0 = const()[name = tensor("op_10034_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10034_cast_fp16 = slice_by_index(begin = var_10034_begin_0, end = var_10034_end_0, end_mask = var_10034_end_mask_0, x = transpose_7)[name = tensor("op_10034_cast_fp16")]; tensor var_10038_begin_0 = const()[name = tensor("op_10038_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_10038_end_0 = const()[name = tensor("op_10038_end_0"), val = tensor([2, 576, 1, 192])]; tensor var_10038_end_mask_0 = const()[name = tensor("op_10038_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10038_cast_fp16 = slice_by_index(begin = var_10038_begin_0, end = var_10038_end_0, end_mask = var_10038_end_mask_0, x = transpose_7)[name = tensor("op_10038_cast_fp16")]; tensor var_10042_begin_0 = const()[name = tensor("op_10042_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_10042_end_0 = const()[name = tensor("op_10042_end_0"), val = tensor([2, 576, 1, 256])]; tensor var_10042_end_mask_0 = const()[name = tensor("op_10042_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10042_cast_fp16 = slice_by_index(begin = var_10042_begin_0, end = var_10042_end_0, end_mask = var_10042_end_mask_0, x = transpose_7)[name = tensor("op_10042_cast_fp16")]; tensor var_10046_begin_0 = const()[name = tensor("op_10046_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_10046_end_0 = const()[name = tensor("op_10046_end_0"), val = tensor([2, 576, 1, 320])]; tensor var_10046_end_mask_0 = const()[name = tensor("op_10046_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10046_cast_fp16 = slice_by_index(begin = var_10046_begin_0, end = var_10046_end_0, end_mask = var_10046_end_mask_0, x = transpose_7)[name = tensor("op_10046_cast_fp16")]; tensor var_10050_begin_0 = const()[name = tensor("op_10050_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_10050_end_0 = const()[name = tensor("op_10050_end_0"), val = tensor([2, 576, 1, 384])]; tensor var_10050_end_mask_0 = const()[name = tensor("op_10050_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10050_cast_fp16 = slice_by_index(begin = var_10050_begin_0, end = var_10050_end_0, end_mask = var_10050_end_mask_0, x = transpose_7)[name = tensor("op_10050_cast_fp16")]; tensor var_10054_begin_0 = const()[name = tensor("op_10054_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_10054_end_0 = const()[name = tensor("op_10054_end_0"), val = tensor([2, 576, 1, 448])]; tensor var_10054_end_mask_0 = const()[name = tensor("op_10054_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10054_cast_fp16 = slice_by_index(begin = var_10054_begin_0, end = var_10054_end_0, end_mask = var_10054_end_mask_0, x = transpose_7)[name = tensor("op_10054_cast_fp16")]; tensor var_10058_begin_0 = const()[name = tensor("op_10058_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_10058_end_0 = const()[name = tensor("op_10058_end_0"), val = tensor([2, 576, 1, 512])]; tensor var_10058_end_mask_0 = const()[name = tensor("op_10058_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10058_cast_fp16 = slice_by_index(begin = var_10058_begin_0, end = var_10058_end_0, end_mask = var_10058_end_mask_0, x = transpose_7)[name = tensor("op_10058_cast_fp16")]; tensor var_10062_begin_0 = const()[name = tensor("op_10062_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10062_end_0 = const()[name = tensor("op_10062_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_10062_end_mask_0 = const()[name = tensor("op_10062_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10062_cast_fp16 = slice_by_index(begin = var_10062_begin_0, end = var_10062_end_0, end_mask = var_10062_end_mask_0, x = transpose_7)[name = tensor("op_10062_cast_fp16")]; tensor var_10066_begin_0 = const()[name = tensor("op_10066_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_10066_end_0 = const()[name = tensor("op_10066_end_0"), val = tensor([2, 576, 1, 640])]; tensor var_10066_end_mask_0 = const()[name = tensor("op_10066_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10066_cast_fp16 = slice_by_index(begin = var_10066_begin_0, end = var_10066_end_0, end_mask = var_10066_end_mask_0, x = transpose_7)[name = tensor("op_10066_cast_fp16")]; tensor var_10068_begin_0 = const()[name = tensor("op_10068_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10068_end_0 = const()[name = tensor("op_10068_end_0"), val = tensor([2, 64, 1, 576])]; tensor var_10068_end_mask_0 = const()[name = tensor("op_10068_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10068_cast_fp16 = slice_by_index(begin = var_10068_begin_0, end = var_10068_end_0, end_mask = var_10068_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10068_cast_fp16")]; tensor var_10072_begin_0 = const()[name = tensor("op_10072_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_10072_end_0 = const()[name = tensor("op_10072_end_0"), val = tensor([2, 128, 1, 576])]; tensor var_10072_end_mask_0 = const()[name = tensor("op_10072_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10072_cast_fp16 = slice_by_index(begin = var_10072_begin_0, end = var_10072_end_0, end_mask = var_10072_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10072_cast_fp16")]; tensor var_10076_begin_0 = const()[name = tensor("op_10076_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_10076_end_0 = const()[name = tensor("op_10076_end_0"), val = tensor([2, 192, 1, 576])]; tensor var_10076_end_mask_0 = const()[name = tensor("op_10076_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10076_cast_fp16 = slice_by_index(begin = var_10076_begin_0, end = var_10076_end_0, end_mask = var_10076_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10076_cast_fp16")]; tensor var_10080_begin_0 = const()[name = tensor("op_10080_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_10080_end_0 = const()[name = tensor("op_10080_end_0"), val = tensor([2, 256, 1, 576])]; tensor var_10080_end_mask_0 = const()[name = tensor("op_10080_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10080_cast_fp16 = slice_by_index(begin = var_10080_begin_0, end = var_10080_end_0, end_mask = var_10080_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10080_cast_fp16")]; tensor var_10084_begin_0 = const()[name = tensor("op_10084_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10084_end_0 = const()[name = tensor("op_10084_end_0"), val = tensor([2, 320, 1, 576])]; tensor var_10084_end_mask_0 = const()[name = tensor("op_10084_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10084_cast_fp16 = slice_by_index(begin = var_10084_begin_0, end = var_10084_end_0, end_mask = var_10084_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10084_cast_fp16")]; tensor var_10088_begin_0 = const()[name = tensor("op_10088_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_10088_end_0 = const()[name = tensor("op_10088_end_0"), val = tensor([2, 384, 1, 576])]; tensor var_10088_end_mask_0 = const()[name = tensor("op_10088_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10088_cast_fp16 = slice_by_index(begin = var_10088_begin_0, end = var_10088_end_0, end_mask = var_10088_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10088_cast_fp16")]; tensor var_10092_begin_0 = const()[name = tensor("op_10092_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_10092_end_0 = const()[name = tensor("op_10092_end_0"), val = tensor([2, 448, 1, 576])]; tensor var_10092_end_mask_0 = const()[name = tensor("op_10092_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10092_cast_fp16 = slice_by_index(begin = var_10092_begin_0, end = var_10092_end_0, end_mask = var_10092_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10092_cast_fp16")]; tensor var_10096_begin_0 = const()[name = tensor("op_10096_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_10096_end_0 = const()[name = tensor("op_10096_end_0"), val = tensor([2, 512, 1, 576])]; tensor var_10096_end_mask_0 = const()[name = tensor("op_10096_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10096_cast_fp16 = slice_by_index(begin = var_10096_begin_0, end = var_10096_end_0, end_mask = var_10096_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10096_cast_fp16")]; tensor var_10100_begin_0 = const()[name = tensor("op_10100_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_10100_end_0 = const()[name = tensor("op_10100_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_10100_end_mask_0 = const()[name = tensor("op_10100_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10100_cast_fp16 = slice_by_index(begin = var_10100_begin_0, end = var_10100_end_0, end_mask = var_10100_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10100_cast_fp16")]; tensor var_10104_begin_0 = const()[name = tensor("op_10104_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_10104_end_0 = const()[name = tensor("op_10104_end_0"), val = tensor([2, 640, 1, 576])]; tensor var_10104_end_mask_0 = const()[name = tensor("op_10104_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10104_cast_fp16 = slice_by_index(begin = var_10104_begin_0, end = var_10104_end_0, end_mask = var_10104_end_mask_0, x = v_49_cast_fp16)[name = tensor("op_10104_cast_fp16")]; tensor var_10108_equation_0 = const()[name = tensor("op_10108_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10108_cast_fp16 = einsum(equation = var_10108_equation_0, values = (var_10030_cast_fp16, var_9987_cast_fp16))[name = tensor("op_10108_cast_fp16")]; tensor var_10109_to_fp16 = const()[name = tensor("op_10109_to_fp16"), val = tensor(0x1p-3)]; tensor aw_681_cast_fp16 = mul(x = var_10108_cast_fp16, y = var_10109_to_fp16)[name = tensor("aw_681_cast_fp16")]; tensor var_10112_equation_0 = const()[name = tensor("op_10112_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10112_cast_fp16 = einsum(equation = var_10112_equation_0, values = (var_10034_cast_fp16, var_9991_cast_fp16))[name = tensor("op_10112_cast_fp16")]; tensor var_10113_to_fp16 = const()[name = tensor("op_10113_to_fp16"), val = tensor(0x1p-3)]; tensor aw_683_cast_fp16 = mul(x = var_10112_cast_fp16, y = var_10113_to_fp16)[name = tensor("aw_683_cast_fp16")]; tensor var_10116_equation_0 = const()[name = tensor("op_10116_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10116_cast_fp16 = einsum(equation = var_10116_equation_0, values = (var_10038_cast_fp16, var_9995_cast_fp16))[name = tensor("op_10116_cast_fp16")]; tensor var_10117_to_fp16 = const()[name = tensor("op_10117_to_fp16"), val = tensor(0x1p-3)]; tensor aw_685_cast_fp16 = mul(x = var_10116_cast_fp16, y = var_10117_to_fp16)[name = tensor("aw_685_cast_fp16")]; tensor var_10120_equation_0 = const()[name = tensor("op_10120_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10120_cast_fp16 = einsum(equation = var_10120_equation_0, values = (var_10042_cast_fp16, var_9999_cast_fp16))[name = tensor("op_10120_cast_fp16")]; tensor var_10121_to_fp16 = const()[name = tensor("op_10121_to_fp16"), val = tensor(0x1p-3)]; tensor aw_687_cast_fp16 = mul(x = var_10120_cast_fp16, y = var_10121_to_fp16)[name = tensor("aw_687_cast_fp16")]; tensor var_10124_equation_0 = const()[name = tensor("op_10124_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10124_cast_fp16 = einsum(equation = var_10124_equation_0, values = (var_10046_cast_fp16, var_10003_cast_fp16))[name = tensor("op_10124_cast_fp16")]; tensor var_10125_to_fp16 = const()[name = tensor("op_10125_to_fp16"), val = tensor(0x1p-3)]; tensor aw_689_cast_fp16 = mul(x = var_10124_cast_fp16, y = var_10125_to_fp16)[name = tensor("aw_689_cast_fp16")]; tensor var_10128_equation_0 = const()[name = tensor("op_10128_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10128_cast_fp16 = einsum(equation = var_10128_equation_0, values = (var_10050_cast_fp16, var_10007_cast_fp16))[name = tensor("op_10128_cast_fp16")]; tensor var_10129_to_fp16 = const()[name = tensor("op_10129_to_fp16"), val = tensor(0x1p-3)]; tensor aw_691_cast_fp16 = mul(x = var_10128_cast_fp16, y = var_10129_to_fp16)[name = tensor("aw_691_cast_fp16")]; tensor var_10132_equation_0 = const()[name = tensor("op_10132_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10132_cast_fp16 = einsum(equation = var_10132_equation_0, values = (var_10054_cast_fp16, var_10011_cast_fp16))[name = tensor("op_10132_cast_fp16")]; tensor var_10133_to_fp16 = const()[name = tensor("op_10133_to_fp16"), val = tensor(0x1p-3)]; tensor aw_693_cast_fp16 = mul(x = var_10132_cast_fp16, y = var_10133_to_fp16)[name = tensor("aw_693_cast_fp16")]; tensor var_10136_equation_0 = const()[name = tensor("op_10136_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10136_cast_fp16 = einsum(equation = var_10136_equation_0, values = (var_10058_cast_fp16, var_10015_cast_fp16))[name = tensor("op_10136_cast_fp16")]; tensor var_10137_to_fp16 = const()[name = tensor("op_10137_to_fp16"), val = tensor(0x1p-3)]; tensor aw_695_cast_fp16 = mul(x = var_10136_cast_fp16, y = var_10137_to_fp16)[name = tensor("aw_695_cast_fp16")]; tensor var_10140_equation_0 = const()[name = tensor("op_10140_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10140_cast_fp16 = einsum(equation = var_10140_equation_0, values = (var_10062_cast_fp16, var_10019_cast_fp16))[name = tensor("op_10140_cast_fp16")]; tensor var_10141_to_fp16 = const()[name = tensor("op_10141_to_fp16"), val = tensor(0x1p-3)]; tensor aw_697_cast_fp16 = mul(x = var_10140_cast_fp16, y = var_10141_to_fp16)[name = tensor("aw_697_cast_fp16")]; tensor var_10144_equation_0 = const()[name = tensor("op_10144_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10144_cast_fp16 = einsum(equation = var_10144_equation_0, values = (var_10066_cast_fp16, var_10023_cast_fp16))[name = tensor("op_10144_cast_fp16")]; tensor var_10145_to_fp16 = const()[name = tensor("op_10145_to_fp16"), val = tensor(0x1p-3)]; tensor aw_699_cast_fp16 = mul(x = var_10144_cast_fp16, y = var_10145_to_fp16)[name = tensor("aw_699_cast_fp16")]; tensor var_10147_cast_fp16 = softmax(axis = var_8627, x = aw_681_cast_fp16)[name = tensor("op_10147_cast_fp16")]; tensor var_10148_cast_fp16 = softmax(axis = var_8627, x = aw_683_cast_fp16)[name = tensor("op_10148_cast_fp16")]; tensor var_10149_cast_fp16 = softmax(axis = var_8627, x = aw_685_cast_fp16)[name = tensor("op_10149_cast_fp16")]; tensor var_10150_cast_fp16 = softmax(axis = var_8627, x = aw_687_cast_fp16)[name = tensor("op_10150_cast_fp16")]; tensor var_10151_cast_fp16 = softmax(axis = var_8627, x = aw_689_cast_fp16)[name = tensor("op_10151_cast_fp16")]; tensor var_10152_cast_fp16 = softmax(axis = var_8627, x = aw_691_cast_fp16)[name = tensor("op_10152_cast_fp16")]; tensor var_10153_cast_fp16 = softmax(axis = var_8627, x = aw_693_cast_fp16)[name = tensor("op_10153_cast_fp16")]; tensor var_10154_cast_fp16 = softmax(axis = var_8627, x = aw_695_cast_fp16)[name = tensor("op_10154_cast_fp16")]; tensor var_10155_cast_fp16 = softmax(axis = var_8627, x = aw_697_cast_fp16)[name = tensor("op_10155_cast_fp16")]; tensor var_10156_cast_fp16 = softmax(axis = var_8627, x = aw_699_cast_fp16)[name = tensor("op_10156_cast_fp16")]; tensor var_10158_equation_0 = const()[name = tensor("op_10158_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10158_cast_fp16 = einsum(equation = var_10158_equation_0, values = (var_10068_cast_fp16, var_10147_cast_fp16))[name = tensor("op_10158_cast_fp16")]; tensor var_10160_equation_0 = const()[name = tensor("op_10160_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10160_cast_fp16 = einsum(equation = var_10160_equation_0, values = (var_10072_cast_fp16, var_10148_cast_fp16))[name = tensor("op_10160_cast_fp16")]; tensor var_10162_equation_0 = const()[name = tensor("op_10162_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10162_cast_fp16 = einsum(equation = var_10162_equation_0, values = (var_10076_cast_fp16, var_10149_cast_fp16))[name = tensor("op_10162_cast_fp16")]; tensor var_10164_equation_0 = const()[name = tensor("op_10164_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10164_cast_fp16 = einsum(equation = var_10164_equation_0, values = (var_10080_cast_fp16, var_10150_cast_fp16))[name = tensor("op_10164_cast_fp16")]; tensor var_10166_equation_0 = const()[name = tensor("op_10166_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10166_cast_fp16 = einsum(equation = var_10166_equation_0, values = (var_10084_cast_fp16, var_10151_cast_fp16))[name = tensor("op_10166_cast_fp16")]; tensor var_10168_equation_0 = const()[name = tensor("op_10168_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10168_cast_fp16 = einsum(equation = var_10168_equation_0, values = (var_10088_cast_fp16, var_10152_cast_fp16))[name = tensor("op_10168_cast_fp16")]; tensor var_10170_equation_0 = const()[name = tensor("op_10170_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10170_cast_fp16 = einsum(equation = var_10170_equation_0, values = (var_10092_cast_fp16, var_10153_cast_fp16))[name = tensor("op_10170_cast_fp16")]; tensor var_10172_equation_0 = const()[name = tensor("op_10172_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10172_cast_fp16 = einsum(equation = var_10172_equation_0, values = (var_10096_cast_fp16, var_10154_cast_fp16))[name = tensor("op_10172_cast_fp16")]; tensor var_10174_equation_0 = const()[name = tensor("op_10174_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10174_cast_fp16 = einsum(equation = var_10174_equation_0, values = (var_10100_cast_fp16, var_10155_cast_fp16))[name = tensor("op_10174_cast_fp16")]; tensor var_10176_equation_0 = const()[name = tensor("op_10176_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10176_cast_fp16 = einsum(equation = var_10176_equation_0, values = (var_10104_cast_fp16, var_10156_cast_fp16))[name = tensor("op_10176_cast_fp16")]; tensor input_435_interleave_0 = const()[name = tensor("input_435_interleave_0"), val = tensor(false)]; tensor input_435_cast_fp16 = concat(axis = var_8627, interleave = input_435_interleave_0, values = (var_10158_cast_fp16, var_10160_cast_fp16, var_10162_cast_fp16, var_10164_cast_fp16, var_10166_cast_fp16, var_10168_cast_fp16, var_10170_cast_fp16, var_10172_cast_fp16, var_10174_cast_fp16, var_10176_cast_fp16))[name = tensor("input_435_cast_fp16")]; tensor var_10182 = const()[name = tensor("op_10182"), val = tensor([1, 1])]; tensor var_10184 = const()[name = tensor("op_10184"), val = tensor([1, 1])]; tensor var_10186_pad_type_0 = const()[name = tensor("op_10186_pad_type_0"), val = tensor("custom")]; tensor var_10186_pad_0 = const()[name = tensor("op_10186_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(776688768)))]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(777508032)))]; tensor var_10186_cast_fp16 = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_10184, groups = var_8627, pad = var_10186_pad_0, pad_type = var_10186_pad_type_0, strides = var_10182, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_435_cast_fp16)[name = tensor("op_10186_cast_fp16")]; tensor inputs_75_cast_fp16 = add(x = var_10186_cast_fp16, y = inputs_73_cast_fp16)[name = tensor("inputs_75_cast_fp16")]; tensor var_10190 = const()[name = tensor("op_10190"), val = tensor([1])]; tensor channels_mean_75_cast_fp16 = reduce_mean(axes = var_10190, keep_dims = var_8622, x = inputs_75_cast_fp16)[name = tensor("channels_mean_75_cast_fp16")]; tensor zero_mean_75_cast_fp16 = sub(x = inputs_75_cast_fp16, y = channels_mean_75_cast_fp16)[name = tensor("zero_mean_75_cast_fp16")]; tensor zero_mean_sq_75_cast_fp16 = mul(x = zero_mean_75_cast_fp16, y = zero_mean_75_cast_fp16)[name = tensor("zero_mean_sq_75_cast_fp16")]; tensor var_10194 = const()[name = tensor("op_10194"), val = tensor([1])]; tensor var_10195_cast_fp16 = reduce_mean(axes = var_10194, keep_dims = var_8622, x = zero_mean_sq_75_cast_fp16)[name = tensor("op_10195_cast_fp16")]; tensor var_10196_to_fp16 = const()[name = tensor("op_10196_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_10197_cast_fp16 = add(x = var_10195_cast_fp16, y = var_10196_to_fp16)[name = tensor("op_10197_cast_fp16")]; tensor denom_75_epsilon_0_to_fp16 = const()[name = tensor("denom_75_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_75_cast_fp16 = rsqrt(epsilon = denom_75_epsilon_0_to_fp16, x = var_10197_cast_fp16)[name = tensor("denom_75_cast_fp16")]; tensor out_75_cast_fp16 = mul(x = zero_mean_75_cast_fp16, y = denom_75_cast_fp16)[name = tensor("out_75_cast_fp16")]; tensor var_10201_to_fp16 = const()[name = tensor("op_10201_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(777509376)))]; tensor var_10202_cast_fp16 = add(x = out_75_cast_fp16, y = var_10201_to_fp16)[name = tensor("op_10202_cast_fp16")]; tensor var_10204_to_fp16 = const()[name = tensor("op_10204_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(777510720)))]; tensor hidden_states_269_cast_fp16 = mul(x = var_10202_cast_fp16, y = var_10204_to_fp16)[name = tensor("hidden_states_269_cast_fp16")]; tensor var_10211 = const()[name = tensor("op_10211"), val = tensor([1, 1])]; tensor var_10213 = const()[name = tensor("op_10213"), val = tensor([1, 1])]; tensor q_51_pad_type_0 = const()[name = tensor("q_51_pad_type_0"), val = tensor("custom")]; tensor q_51_pad_0 = const()[name = tensor("q_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(777512064)))]; tensor q_51_cast_fp16 = conv(dilations = var_10213, groups = var_8627, pad = q_51_pad_0, pad_type = q_51_pad_type_0, strides = var_10211, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_269_cast_fp16)[name = tensor("q_51_cast_fp16")]; tensor var_10217 = const()[name = tensor("op_10217"), val = tensor([1, 1])]; tensor var_10219 = const()[name = tensor("op_10219"), val = tensor([1, 1])]; tensor k_101_pad_type_0 = const()[name = tensor("k_101_pad_type_0"), val = tensor("custom")]; tensor k_101_pad_0 = const()[name = tensor("k_101_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(778331328)))]; tensor k_101_cast_fp16 = conv(dilations = var_10219, groups = var_8627, pad = k_101_pad_0, pad_type = k_101_pad_type_0, strides = var_10217, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_101_cast_fp16")]; tensor var_10223 = const()[name = tensor("op_10223"), val = tensor([1, 1])]; tensor var_10225 = const()[name = tensor("op_10225"), val = tensor([1, 1])]; tensor v_51_pad_type_0 = const()[name = tensor("v_51_pad_type_0"), val = tensor("custom")]; tensor v_51_pad_0 = const()[name = tensor("v_51_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(779642112)))]; tensor v_51_cast_fp16 = conv(dilations = var_10225, groups = var_8627, pad = v_51_pad_0, pad_type = v_51_pad_type_0, strides = var_10223, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_51_cast_fp16")]; tensor var_10229_begin_0 = const()[name = tensor("op_10229_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10229_end_0 = const()[name = tensor("op_10229_end_0"), val = tensor([2, 64, 1, 576])]; tensor var_10229_end_mask_0 = const()[name = tensor("op_10229_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10229_cast_fp16 = slice_by_index(begin = var_10229_begin_0, end = var_10229_end_0, end_mask = var_10229_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10229_cast_fp16")]; tensor var_10233_begin_0 = const()[name = tensor("op_10233_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_10233_end_0 = const()[name = tensor("op_10233_end_0"), val = tensor([2, 128, 1, 576])]; tensor var_10233_end_mask_0 = const()[name = tensor("op_10233_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10233_cast_fp16 = slice_by_index(begin = var_10233_begin_0, end = var_10233_end_0, end_mask = var_10233_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10233_cast_fp16")]; tensor var_10237_begin_0 = const()[name = tensor("op_10237_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_10237_end_0 = const()[name = tensor("op_10237_end_0"), val = tensor([2, 192, 1, 576])]; tensor var_10237_end_mask_0 = const()[name = tensor("op_10237_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10237_cast_fp16 = slice_by_index(begin = var_10237_begin_0, end = var_10237_end_0, end_mask = var_10237_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10237_cast_fp16")]; tensor var_10241_begin_0 = const()[name = tensor("op_10241_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_10241_end_0 = const()[name = tensor("op_10241_end_0"), val = tensor([2, 256, 1, 576])]; tensor var_10241_end_mask_0 = const()[name = tensor("op_10241_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10241_cast_fp16 = slice_by_index(begin = var_10241_begin_0, end = var_10241_end_0, end_mask = var_10241_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10241_cast_fp16")]; tensor var_10245_begin_0 = const()[name = tensor("op_10245_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10245_end_0 = const()[name = tensor("op_10245_end_0"), val = tensor([2, 320, 1, 576])]; tensor var_10245_end_mask_0 = const()[name = tensor("op_10245_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10245_cast_fp16 = slice_by_index(begin = var_10245_begin_0, end = var_10245_end_0, end_mask = var_10245_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10245_cast_fp16")]; tensor var_10249_begin_0 = const()[name = tensor("op_10249_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_10249_end_0 = const()[name = tensor("op_10249_end_0"), val = tensor([2, 384, 1, 576])]; tensor var_10249_end_mask_0 = const()[name = tensor("op_10249_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10249_cast_fp16 = slice_by_index(begin = var_10249_begin_0, end = var_10249_end_0, end_mask = var_10249_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10249_cast_fp16")]; tensor var_10253_begin_0 = const()[name = tensor("op_10253_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_10253_end_0 = const()[name = tensor("op_10253_end_0"), val = tensor([2, 448, 1, 576])]; tensor var_10253_end_mask_0 = const()[name = tensor("op_10253_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10253_cast_fp16 = slice_by_index(begin = var_10253_begin_0, end = var_10253_end_0, end_mask = var_10253_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10253_cast_fp16")]; tensor var_10257_begin_0 = const()[name = tensor("op_10257_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_10257_end_0 = const()[name = tensor("op_10257_end_0"), val = tensor([2, 512, 1, 576])]; tensor var_10257_end_mask_0 = const()[name = tensor("op_10257_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10257_cast_fp16 = slice_by_index(begin = var_10257_begin_0, end = var_10257_end_0, end_mask = var_10257_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10257_cast_fp16")]; tensor var_10261_begin_0 = const()[name = tensor("op_10261_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_10261_end_0 = const()[name = tensor("op_10261_end_0"), val = tensor([2, 576, 1, 576])]; tensor var_10261_end_mask_0 = const()[name = tensor("op_10261_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10261_cast_fp16 = slice_by_index(begin = var_10261_begin_0, end = var_10261_end_0, end_mask = var_10261_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10261_cast_fp16")]; tensor var_10265_begin_0 = const()[name = tensor("op_10265_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_10265_end_0 = const()[name = tensor("op_10265_end_0"), val = tensor([2, 640, 1, 576])]; tensor var_10265_end_mask_0 = const()[name = tensor("op_10265_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10265_cast_fp16 = slice_by_index(begin = var_10265_begin_0, end = var_10265_end_0, end_mask = var_10265_end_mask_0, x = q_51_cast_fp16)[name = tensor("op_10265_cast_fp16")]; tensor k_103_perm_0 = const()[name = tensor("k_103_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_10272_begin_0 = const()[name = tensor("op_10272_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10272_end_0 = const()[name = tensor("op_10272_end_0"), val = tensor([2, 77, 1, 64])]; tensor var_10272_end_mask_0 = const()[name = tensor("op_10272_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_6 = transpose(perm = k_103_perm_0, x = k_101_cast_fp16)[name = tensor("transpose_6")]; tensor var_10272_cast_fp16 = slice_by_index(begin = var_10272_begin_0, end = var_10272_end_0, end_mask = var_10272_end_mask_0, x = transpose_6)[name = tensor("op_10272_cast_fp16")]; tensor var_10276_begin_0 = const()[name = tensor("op_10276_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_10276_end_0 = const()[name = tensor("op_10276_end_0"), val = tensor([2, 77, 1, 128])]; tensor var_10276_end_mask_0 = const()[name = tensor("op_10276_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10276_cast_fp16 = slice_by_index(begin = var_10276_begin_0, end = var_10276_end_0, end_mask = var_10276_end_mask_0, x = transpose_6)[name = tensor("op_10276_cast_fp16")]; tensor var_10280_begin_0 = const()[name = tensor("op_10280_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_10280_end_0 = const()[name = tensor("op_10280_end_0"), val = tensor([2, 77, 1, 192])]; tensor var_10280_end_mask_0 = const()[name = tensor("op_10280_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10280_cast_fp16 = slice_by_index(begin = var_10280_begin_0, end = var_10280_end_0, end_mask = var_10280_end_mask_0, x = transpose_6)[name = tensor("op_10280_cast_fp16")]; tensor var_10284_begin_0 = const()[name = tensor("op_10284_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_10284_end_0 = const()[name = tensor("op_10284_end_0"), val = tensor([2, 77, 1, 256])]; tensor var_10284_end_mask_0 = const()[name = tensor("op_10284_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10284_cast_fp16 = slice_by_index(begin = var_10284_begin_0, end = var_10284_end_0, end_mask = var_10284_end_mask_0, x = transpose_6)[name = tensor("op_10284_cast_fp16")]; tensor var_10288_begin_0 = const()[name = tensor("op_10288_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_10288_end_0 = const()[name = tensor("op_10288_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_10288_end_mask_0 = const()[name = tensor("op_10288_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10288_cast_fp16 = slice_by_index(begin = var_10288_begin_0, end = var_10288_end_0, end_mask = var_10288_end_mask_0, x = transpose_6)[name = tensor("op_10288_cast_fp16")]; tensor var_10292_begin_0 = const()[name = tensor("op_10292_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_10292_end_0 = const()[name = tensor("op_10292_end_0"), val = tensor([2, 77, 1, 384])]; tensor var_10292_end_mask_0 = const()[name = tensor("op_10292_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10292_cast_fp16 = slice_by_index(begin = var_10292_begin_0, end = var_10292_end_0, end_mask = var_10292_end_mask_0, x = transpose_6)[name = tensor("op_10292_cast_fp16")]; tensor var_10296_begin_0 = const()[name = tensor("op_10296_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_10296_end_0 = const()[name = tensor("op_10296_end_0"), val = tensor([2, 77, 1, 448])]; tensor var_10296_end_mask_0 = const()[name = tensor("op_10296_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10296_cast_fp16 = slice_by_index(begin = var_10296_begin_0, end = var_10296_end_0, end_mask = var_10296_end_mask_0, x = transpose_6)[name = tensor("op_10296_cast_fp16")]; tensor var_10300_begin_0 = const()[name = tensor("op_10300_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_10300_end_0 = const()[name = tensor("op_10300_end_0"), val = tensor([2, 77, 1, 512])]; tensor var_10300_end_mask_0 = const()[name = tensor("op_10300_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10300_cast_fp16 = slice_by_index(begin = var_10300_begin_0, end = var_10300_end_0, end_mask = var_10300_end_mask_0, x = transpose_6)[name = tensor("op_10300_cast_fp16")]; tensor var_10304_begin_0 = const()[name = tensor("op_10304_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10304_end_0 = const()[name = tensor("op_10304_end_0"), val = tensor([2, 77, 1, 576])]; tensor var_10304_end_mask_0 = const()[name = tensor("op_10304_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10304_cast_fp16 = slice_by_index(begin = var_10304_begin_0, end = var_10304_end_0, end_mask = var_10304_end_mask_0, x = transpose_6)[name = tensor("op_10304_cast_fp16")]; tensor var_10308_begin_0 = const()[name = tensor("op_10308_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_10308_end_0 = const()[name = tensor("op_10308_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_10308_end_mask_0 = const()[name = tensor("op_10308_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10308_cast_fp16 = slice_by_index(begin = var_10308_begin_0, end = var_10308_end_0, end_mask = var_10308_end_mask_0, x = transpose_6)[name = tensor("op_10308_cast_fp16")]; tensor var_10310_begin_0 = const()[name = tensor("op_10310_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10310_end_0 = const()[name = tensor("op_10310_end_0"), val = tensor([2, 64, 1, 77])]; tensor var_10310_end_mask_0 = const()[name = tensor("op_10310_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10310_cast_fp16 = slice_by_index(begin = var_10310_begin_0, end = var_10310_end_0, end_mask = var_10310_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10310_cast_fp16")]; tensor var_10314_begin_0 = const()[name = tensor("op_10314_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_10314_end_0 = const()[name = tensor("op_10314_end_0"), val = tensor([2, 128, 1, 77])]; tensor var_10314_end_mask_0 = const()[name = tensor("op_10314_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10314_cast_fp16 = slice_by_index(begin = var_10314_begin_0, end = var_10314_end_0, end_mask = var_10314_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10314_cast_fp16")]; tensor var_10318_begin_0 = const()[name = tensor("op_10318_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_10318_end_0 = const()[name = tensor("op_10318_end_0"), val = tensor([2, 192, 1, 77])]; tensor var_10318_end_mask_0 = const()[name = tensor("op_10318_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10318_cast_fp16 = slice_by_index(begin = var_10318_begin_0, end = var_10318_end_0, end_mask = var_10318_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10318_cast_fp16")]; tensor var_10322_begin_0 = const()[name = tensor("op_10322_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_10322_end_0 = const()[name = tensor("op_10322_end_0"), val = tensor([2, 256, 1, 77])]; tensor var_10322_end_mask_0 = const()[name = tensor("op_10322_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10322_cast_fp16 = slice_by_index(begin = var_10322_begin_0, end = var_10322_end_0, end_mask = var_10322_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10322_cast_fp16")]; tensor var_10326_begin_0 = const()[name = tensor("op_10326_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10326_end_0 = const()[name = tensor("op_10326_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_10326_end_mask_0 = const()[name = tensor("op_10326_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10326_cast_fp16 = slice_by_index(begin = var_10326_begin_0, end = var_10326_end_0, end_mask = var_10326_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10326_cast_fp16")]; tensor var_10330_begin_0 = const()[name = tensor("op_10330_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_10330_end_0 = const()[name = tensor("op_10330_end_0"), val = tensor([2, 384, 1, 77])]; tensor var_10330_end_mask_0 = const()[name = tensor("op_10330_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10330_cast_fp16 = slice_by_index(begin = var_10330_begin_0, end = var_10330_end_0, end_mask = var_10330_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10330_cast_fp16")]; tensor var_10334_begin_0 = const()[name = tensor("op_10334_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_10334_end_0 = const()[name = tensor("op_10334_end_0"), val = tensor([2, 448, 1, 77])]; tensor var_10334_end_mask_0 = const()[name = tensor("op_10334_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10334_cast_fp16 = slice_by_index(begin = var_10334_begin_0, end = var_10334_end_0, end_mask = var_10334_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10334_cast_fp16")]; tensor var_10338_begin_0 = const()[name = tensor("op_10338_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_10338_end_0 = const()[name = tensor("op_10338_end_0"), val = tensor([2, 512, 1, 77])]; tensor var_10338_end_mask_0 = const()[name = tensor("op_10338_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10338_cast_fp16 = slice_by_index(begin = var_10338_begin_0, end = var_10338_end_0, end_mask = var_10338_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10338_cast_fp16")]; tensor var_10342_begin_0 = const()[name = tensor("op_10342_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_10342_end_0 = const()[name = tensor("op_10342_end_0"), val = tensor([2, 576, 1, 77])]; tensor var_10342_end_mask_0 = const()[name = tensor("op_10342_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10342_cast_fp16 = slice_by_index(begin = var_10342_begin_0, end = var_10342_end_0, end_mask = var_10342_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10342_cast_fp16")]; tensor var_10346_begin_0 = const()[name = tensor("op_10346_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_10346_end_0 = const()[name = tensor("op_10346_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_10346_end_mask_0 = const()[name = tensor("op_10346_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10346_cast_fp16 = slice_by_index(begin = var_10346_begin_0, end = var_10346_end_0, end_mask = var_10346_end_mask_0, x = v_51_cast_fp16)[name = tensor("op_10346_cast_fp16")]; tensor var_10350_equation_0 = const()[name = tensor("op_10350_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10350_cast_fp16 = einsum(equation = var_10350_equation_0, values = (var_10272_cast_fp16, var_10229_cast_fp16))[name = tensor("op_10350_cast_fp16")]; tensor var_10351_to_fp16 = const()[name = tensor("op_10351_to_fp16"), val = tensor(0x1p-3)]; tensor aw_701_cast_fp16 = mul(x = var_10350_cast_fp16, y = var_10351_to_fp16)[name = tensor("aw_701_cast_fp16")]; tensor var_10354_equation_0 = const()[name = tensor("op_10354_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10354_cast_fp16 = einsum(equation = var_10354_equation_0, values = (var_10276_cast_fp16, var_10233_cast_fp16))[name = tensor("op_10354_cast_fp16")]; tensor var_10355_to_fp16 = const()[name = tensor("op_10355_to_fp16"), val = tensor(0x1p-3)]; tensor aw_703_cast_fp16 = mul(x = var_10354_cast_fp16, y = var_10355_to_fp16)[name = tensor("aw_703_cast_fp16")]; tensor var_10358_equation_0 = const()[name = tensor("op_10358_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10358_cast_fp16 = einsum(equation = var_10358_equation_0, values = (var_10280_cast_fp16, var_10237_cast_fp16))[name = tensor("op_10358_cast_fp16")]; tensor var_10359_to_fp16 = const()[name = tensor("op_10359_to_fp16"), val = tensor(0x1p-3)]; tensor aw_705_cast_fp16 = mul(x = var_10358_cast_fp16, y = var_10359_to_fp16)[name = tensor("aw_705_cast_fp16")]; tensor var_10362_equation_0 = const()[name = tensor("op_10362_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10362_cast_fp16 = einsum(equation = var_10362_equation_0, values = (var_10284_cast_fp16, var_10241_cast_fp16))[name = tensor("op_10362_cast_fp16")]; tensor var_10363_to_fp16 = const()[name = tensor("op_10363_to_fp16"), val = tensor(0x1p-3)]; tensor aw_707_cast_fp16 = mul(x = var_10362_cast_fp16, y = var_10363_to_fp16)[name = tensor("aw_707_cast_fp16")]; tensor var_10366_equation_0 = const()[name = tensor("op_10366_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10366_cast_fp16 = einsum(equation = var_10366_equation_0, values = (var_10288_cast_fp16, var_10245_cast_fp16))[name = tensor("op_10366_cast_fp16")]; tensor var_10367_to_fp16 = const()[name = tensor("op_10367_to_fp16"), val = tensor(0x1p-3)]; tensor aw_709_cast_fp16 = mul(x = var_10366_cast_fp16, y = var_10367_to_fp16)[name = tensor("aw_709_cast_fp16")]; tensor var_10370_equation_0 = const()[name = tensor("op_10370_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10370_cast_fp16 = einsum(equation = var_10370_equation_0, values = (var_10292_cast_fp16, var_10249_cast_fp16))[name = tensor("op_10370_cast_fp16")]; tensor var_10371_to_fp16 = const()[name = tensor("op_10371_to_fp16"), val = tensor(0x1p-3)]; tensor aw_711_cast_fp16 = mul(x = var_10370_cast_fp16, y = var_10371_to_fp16)[name = tensor("aw_711_cast_fp16")]; tensor var_10374_equation_0 = const()[name = tensor("op_10374_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10374_cast_fp16 = einsum(equation = var_10374_equation_0, values = (var_10296_cast_fp16, var_10253_cast_fp16))[name = tensor("op_10374_cast_fp16")]; tensor var_10375_to_fp16 = const()[name = tensor("op_10375_to_fp16"), val = tensor(0x1p-3)]; tensor aw_713_cast_fp16 = mul(x = var_10374_cast_fp16, y = var_10375_to_fp16)[name = tensor("aw_713_cast_fp16")]; tensor var_10378_equation_0 = const()[name = tensor("op_10378_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10378_cast_fp16 = einsum(equation = var_10378_equation_0, values = (var_10300_cast_fp16, var_10257_cast_fp16))[name = tensor("op_10378_cast_fp16")]; tensor var_10379_to_fp16 = const()[name = tensor("op_10379_to_fp16"), val = tensor(0x1p-3)]; tensor aw_715_cast_fp16 = mul(x = var_10378_cast_fp16, y = var_10379_to_fp16)[name = tensor("aw_715_cast_fp16")]; tensor var_10382_equation_0 = const()[name = tensor("op_10382_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10382_cast_fp16 = einsum(equation = var_10382_equation_0, values = (var_10304_cast_fp16, var_10261_cast_fp16))[name = tensor("op_10382_cast_fp16")]; tensor var_10383_to_fp16 = const()[name = tensor("op_10383_to_fp16"), val = tensor(0x1p-3)]; tensor aw_717_cast_fp16 = mul(x = var_10382_cast_fp16, y = var_10383_to_fp16)[name = tensor("aw_717_cast_fp16")]; tensor var_10386_equation_0 = const()[name = tensor("op_10386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10386_cast_fp16 = einsum(equation = var_10386_equation_0, values = (var_10308_cast_fp16, var_10265_cast_fp16))[name = tensor("op_10386_cast_fp16")]; tensor var_10387_to_fp16 = const()[name = tensor("op_10387_to_fp16"), val = tensor(0x1p-3)]; tensor aw_719_cast_fp16 = mul(x = var_10386_cast_fp16, y = var_10387_to_fp16)[name = tensor("aw_719_cast_fp16")]; tensor var_10389_cast_fp16 = softmax(axis = var_8627, x = aw_701_cast_fp16)[name = tensor("op_10389_cast_fp16")]; tensor var_10390_cast_fp16 = softmax(axis = var_8627, x = aw_703_cast_fp16)[name = tensor("op_10390_cast_fp16")]; tensor var_10391_cast_fp16 = softmax(axis = var_8627, x = aw_705_cast_fp16)[name = tensor("op_10391_cast_fp16")]; tensor var_10392_cast_fp16 = softmax(axis = var_8627, x = aw_707_cast_fp16)[name = tensor("op_10392_cast_fp16")]; tensor var_10393_cast_fp16 = softmax(axis = var_8627, x = aw_709_cast_fp16)[name = tensor("op_10393_cast_fp16")]; tensor var_10394_cast_fp16 = softmax(axis = var_8627, x = aw_711_cast_fp16)[name = tensor("op_10394_cast_fp16")]; tensor var_10395_cast_fp16 = softmax(axis = var_8627, x = aw_713_cast_fp16)[name = tensor("op_10395_cast_fp16")]; tensor var_10396_cast_fp16 = softmax(axis = var_8627, x = aw_715_cast_fp16)[name = tensor("op_10396_cast_fp16")]; tensor var_10397_cast_fp16 = softmax(axis = var_8627, x = aw_717_cast_fp16)[name = tensor("op_10397_cast_fp16")]; tensor var_10398_cast_fp16 = softmax(axis = var_8627, x = aw_719_cast_fp16)[name = tensor("op_10398_cast_fp16")]; tensor var_10400_equation_0 = const()[name = tensor("op_10400_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10400_cast_fp16 = einsum(equation = var_10400_equation_0, values = (var_10310_cast_fp16, var_10389_cast_fp16))[name = tensor("op_10400_cast_fp16")]; tensor var_10402_equation_0 = const()[name = tensor("op_10402_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10402_cast_fp16 = einsum(equation = var_10402_equation_0, values = (var_10314_cast_fp16, var_10390_cast_fp16))[name = tensor("op_10402_cast_fp16")]; tensor var_10404_equation_0 = const()[name = tensor("op_10404_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10404_cast_fp16 = einsum(equation = var_10404_equation_0, values = (var_10318_cast_fp16, var_10391_cast_fp16))[name = tensor("op_10404_cast_fp16")]; tensor var_10406_equation_0 = const()[name = tensor("op_10406_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10406_cast_fp16 = einsum(equation = var_10406_equation_0, values = (var_10322_cast_fp16, var_10392_cast_fp16))[name = tensor("op_10406_cast_fp16")]; tensor var_10408_equation_0 = const()[name = tensor("op_10408_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10408_cast_fp16 = einsum(equation = var_10408_equation_0, values = (var_10326_cast_fp16, var_10393_cast_fp16))[name = tensor("op_10408_cast_fp16")]; tensor var_10410_equation_0 = const()[name = tensor("op_10410_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10410_cast_fp16 = einsum(equation = var_10410_equation_0, values = (var_10330_cast_fp16, var_10394_cast_fp16))[name = tensor("op_10410_cast_fp16")]; tensor var_10412_equation_0 = const()[name = tensor("op_10412_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10412_cast_fp16 = einsum(equation = var_10412_equation_0, values = (var_10334_cast_fp16, var_10395_cast_fp16))[name = tensor("op_10412_cast_fp16")]; tensor var_10414_equation_0 = const()[name = tensor("op_10414_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10414_cast_fp16 = einsum(equation = var_10414_equation_0, values = (var_10338_cast_fp16, var_10396_cast_fp16))[name = tensor("op_10414_cast_fp16")]; tensor var_10416_equation_0 = const()[name = tensor("op_10416_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10416_cast_fp16 = einsum(equation = var_10416_equation_0, values = (var_10342_cast_fp16, var_10397_cast_fp16))[name = tensor("op_10416_cast_fp16")]; tensor var_10418_equation_0 = const()[name = tensor("op_10418_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10418_cast_fp16 = einsum(equation = var_10418_equation_0, values = (var_10346_cast_fp16, var_10398_cast_fp16))[name = tensor("op_10418_cast_fp16")]; tensor input_437_interleave_0 = const()[name = tensor("input_437_interleave_0"), val = tensor(false)]; tensor input_437_cast_fp16 = concat(axis = var_8627, interleave = input_437_interleave_0, values = (var_10400_cast_fp16, var_10402_cast_fp16, var_10404_cast_fp16, var_10406_cast_fp16, var_10408_cast_fp16, var_10410_cast_fp16, var_10412_cast_fp16, var_10414_cast_fp16, var_10416_cast_fp16, var_10418_cast_fp16))[name = tensor("input_437_cast_fp16")]; tensor var_10424 = const()[name = tensor("op_10424"), val = tensor([1, 1])]; tensor var_10426 = const()[name = tensor("op_10426"), val = tensor([1, 1])]; tensor var_10428_pad_type_0 = const()[name = tensor("op_10428_pad_type_0"), val = tensor("custom")]; tensor var_10428_pad_0 = const()[name = tensor("op_10428_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(780952896)))]; tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(781772160)))]; tensor var_10428_cast_fp16 = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_10426, groups = var_8627, pad = var_10428_pad_0, pad_type = var_10428_pad_type_0, strides = var_10424, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_437_cast_fp16)[name = tensor("op_10428_cast_fp16")]; tensor inputs_77_cast_fp16 = add(x = var_10428_cast_fp16, y = inputs_75_cast_fp16)[name = tensor("inputs_77_cast_fp16")]; tensor var_10432 = const()[name = tensor("op_10432"), val = tensor([1])]; tensor channels_mean_77_cast_fp16 = reduce_mean(axes = var_10432, keep_dims = var_8622, x = inputs_77_cast_fp16)[name = tensor("channels_mean_77_cast_fp16")]; tensor zero_mean_77_cast_fp16 = sub(x = inputs_77_cast_fp16, y = channels_mean_77_cast_fp16)[name = tensor("zero_mean_77_cast_fp16")]; tensor zero_mean_sq_77_cast_fp16 = mul(x = zero_mean_77_cast_fp16, y = zero_mean_77_cast_fp16)[name = tensor("zero_mean_sq_77_cast_fp16")]; tensor var_10436 = const()[name = tensor("op_10436"), val = tensor([1])]; tensor var_10437_cast_fp16 = reduce_mean(axes = var_10436, keep_dims = var_8622, x = zero_mean_sq_77_cast_fp16)[name = tensor("op_10437_cast_fp16")]; tensor var_10438_to_fp16 = const()[name = tensor("op_10438_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_10439_cast_fp16 = add(x = var_10437_cast_fp16, y = var_10438_to_fp16)[name = tensor("op_10439_cast_fp16")]; tensor denom_77_epsilon_0_to_fp16 = const()[name = tensor("denom_77_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_77_cast_fp16 = rsqrt(epsilon = denom_77_epsilon_0_to_fp16, x = var_10439_cast_fp16)[name = tensor("denom_77_cast_fp16")]; tensor out_77_cast_fp16 = mul(x = zero_mean_77_cast_fp16, y = denom_77_cast_fp16)[name = tensor("out_77_cast_fp16")]; tensor var_10443_to_fp16 = const()[name = tensor("op_10443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(781773504)))]; tensor var_10444_cast_fp16 = add(x = out_77_cast_fp16, y = var_10443_to_fp16)[name = tensor("op_10444_cast_fp16")]; tensor var_10446_to_fp16 = const()[name = tensor("op_10446_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(781774848)))]; tensor input_439_cast_fp16 = mul(x = var_10444_cast_fp16, y = var_10446_to_fp16)[name = tensor("input_439_cast_fp16")]; tensor var_10454 = const()[name = tensor("op_10454"), val = tensor([1, 1])]; tensor var_10456 = const()[name = tensor("op_10456"), val = tensor([1, 1])]; tensor var_10458_pad_type_0 = const()[name = tensor("op_10458_pad_type_0"), val = tensor("custom")]; tensor var_10458_pad_0 = const()[name = tensor("op_10458_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(781776192)))]; tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788329856)))]; tensor var_10458_cast_fp16 = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_10456, groups = var_8627, pad = var_10458_pad_0, pad_type = var_10458_pad_type_0, strides = var_10454, weight = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_439_cast_fp16)[name = tensor("op_10458_cast_fp16")]; tensor var_10459_split_sizes_0 = const()[name = tensor("op_10459_split_sizes_0"), val = tensor([2560, 2560])]; tensor var_10459_axis_0 = const()[name = tensor("op_10459_axis_0"), val = tensor(1)]; tensor var_10459_cast_fp16_0, tensor var_10459_cast_fp16_1 = split(axis = var_10459_axis_0, split_sizes = var_10459_split_sizes_0, x = var_10458_cast_fp16)[name = tensor("op_10459_cast_fp16")]; tensor var_10461_mode_0 = const()[name = tensor("op_10461_mode_0"), val = tensor("EXACT")]; tensor var_10461_cast_fp16 = gelu(mode = var_10461_mode_0, x = var_10459_cast_fp16_1)[name = tensor("op_10461_cast_fp16")]; tensor input_441_cast_fp16 = mul(x = var_10459_cast_fp16_0, y = var_10461_cast_fp16)[name = tensor("input_441_cast_fp16")]; tensor var_10465 = const()[name = tensor("op_10465"), val = tensor([1, 1])]; tensor var_10467 = const()[name = tensor("op_10467"), val = tensor([1, 1])]; tensor var_10469_pad_type_0 = const()[name = tensor("op_10469_pad_type_0"), val = tensor("custom")]; tensor var_10469_pad_0 = const()[name = tensor("op_10469_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788340160)))]; tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(791617024)))]; tensor var_10469_cast_fp16 = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_10467, groups = var_8627, pad = var_10469_pad_0, pad_type = var_10469_pad_type_0, strides = var_10465, weight = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_441_cast_fp16)[name = tensor("op_10469_cast_fp16")]; tensor hidden_states_273_cast_fp16 = add(x = var_10469_cast_fp16, y = inputs_77_cast_fp16)[name = tensor("hidden_states_273_cast_fp16")]; tensor var_10471 = const()[name = tensor("op_10471"), val = tensor([2, 640, 24, 24])]; tensor input_443_cast_fp16 = reshape(shape = var_10471, x = hidden_states_273_cast_fp16)[name = tensor("input_443_cast_fp16")]; tensor var_10475 = const()[name = tensor("op_10475"), val = tensor([1, 1])]; tensor var_10477 = const()[name = tensor("op_10477"), val = tensor([1, 1])]; tensor hidden_states_275_pad_type_0 = const()[name = tensor("hidden_states_275_pad_type_0"), val = tensor("custom")]; tensor hidden_states_275_pad_0 = const()[name = tensor("hidden_states_275_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_2_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(791618368)))]; tensor up_blocks_2_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(792437632)))]; tensor hidden_states_275_cast_fp16 = conv(bias = up_blocks_2_attentions_2_proj_out_bias_to_fp16, dilations = var_10477, groups = var_8627, pad = hidden_states_275_pad_0, pad_type = hidden_states_275_pad_type_0, strides = var_10475, weight = up_blocks_2_attentions_2_proj_out_weight_to_fp16, x = input_443_cast_fp16)[name = tensor("hidden_states_275_cast_fp16")]; tensor input_445_cast_fp16 = add(x = hidden_states_275_cast_fp16, y = hidden_states_263_cast_fp16)[name = tensor("input_445_cast_fp16")]; tensor input_447_scale_factor_height_0 = const()[name = tensor("input_447_scale_factor_height_0"), val = tensor(0x1p+1)]; tensor input_447_scale_factor_width_0 = const()[name = tensor("input_447_scale_factor_width_0"), val = tensor(0x1p+1)]; tensor input_447_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_447_scale_factor_height_0, scale_factor_width = input_447_scale_factor_width_0, x = input_445_cast_fp16)[name = tensor("input_447_cast_fp16")]; tensor var_10486 = const()[name = tensor("op_10486"), val = tensor([1, 1])]; tensor var_10488 = const()[name = tensor("op_10488"), val = tensor([1, 1])]; tensor hidden_states_277_pad_type_0 = const()[name = tensor("hidden_states_277_pad_type_0"), val = tensor("custom")]; tensor hidden_states_277_pad_0 = const()[name = tensor("hidden_states_277_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_2_upsamplers_0_conv_weight_to_fp16 = const()[name = tensor("up_blocks_2_upsamplers_0_conv_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(792438976)))]; tensor up_blocks_2_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor("up_blocks_2_upsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(799811840)))]; tensor hidden_states_277_cast_fp16 = conv(bias = up_blocks_2_upsamplers_0_conv_bias_to_fp16, dilations = var_10488, groups = var_8627, pad = hidden_states_277_pad_0, pad_type = hidden_states_277_pad_type_0, strides = var_10486, weight = up_blocks_2_upsamplers_0_conv_weight_to_fp16, x = input_447_cast_fp16)[name = tensor("hidden_states_277_cast_fp16")]; tensor var_10504 = const()[name = tensor("op_10504"), val = tensor(true)]; tensor var_10509 = const()[name = tensor("op_10509"), val = tensor(1)]; tensor input_449_interleave_0 = const()[name = tensor("input_449_interleave_0"), val = tensor(false)]; tensor cast_10 = cast(dtype = cast_5_dtype_0, x = input_61_cast_fp16)[name = tensor("cast_10")]; tensor input_449_cast_fp16 = concat(axis = var_10509, interleave = input_449_interleave_0, values = (hidden_states_277_cast_fp16, cast_10))[name = tensor("input_449_cast_fp16")]; tensor reshape_204_shape_0 = const()[name = tensor("reshape_204_shape_0"), val = tensor([2, 32, 30, 48, 48])]; tensor reshape_204_cast_fp16 = reshape(shape = reshape_204_shape_0, x = input_449_cast_fp16)[name = tensor("reshape_204_cast_fp16")]; tensor reduce_mean_153_axes_0 = const()[name = tensor("reduce_mean_153_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_153_keep_dims_0 = const()[name = tensor("reduce_mean_153_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_153_cast_fp16 = reduce_mean(axes = reduce_mean_153_axes_0, keep_dims = reduce_mean_153_keep_dims_0, x = reshape_204_cast_fp16)[name = tensor("reduce_mean_153_cast_fp16")]; tensor sub_102_cast_fp16 = sub(x = reshape_204_cast_fp16, y = reduce_mean_153_cast_fp16)[name = tensor("sub_102_cast_fp16")]; tensor square_51_cast_fp16 = square(x = sub_102_cast_fp16)[name = tensor("square_51_cast_fp16")]; tensor reduce_mean_155_axes_0 = const()[name = tensor("reduce_mean_155_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_155_keep_dims_0 = const()[name = tensor("reduce_mean_155_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_155_cast_fp16 = reduce_mean(axes = reduce_mean_155_axes_0, keep_dims = reduce_mean_155_keep_dims_0, x = square_51_cast_fp16)[name = tensor("reduce_mean_155_cast_fp16")]; tensor add_102_y_0_to_fp16 = const()[name = tensor("add_102_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_102_cast_fp16 = add(x = reduce_mean_155_cast_fp16, y = add_102_y_0_to_fp16)[name = tensor("add_102_cast_fp16")]; tensor sqrt_51_cast_fp16 = sqrt(x = add_102_cast_fp16)[name = tensor("sqrt_51_cast_fp16")]; tensor real_div_51_cast_fp16 = real_div(x = sub_102_cast_fp16, y = sqrt_51_cast_fp16)[name = tensor("real_div_51_cast_fp16")]; tensor reshape_205_shape_0 = const()[name = tensor("reshape_205_shape_0"), val = tensor([2, 960, 48, 48])]; tensor reshape_205_cast_fp16 = reshape(shape = reshape_205_shape_0, x = real_div_51_cast_fp16)[name = tensor("reshape_205_cast_fp16")]; tensor add_103_gamma_0_to_fp16 = const()[name = tensor("add_103_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(799813184)))]; tensor add_103_beta_0_to_fp16 = const()[name = tensor("add_103_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(799815168)))]; tensor add_103_epsilon_0_to_fp16 = const()[name = tensor("add_103_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_103_cast_fp16 = batch_norm(beta = add_103_beta_0_to_fp16, epsilon = add_103_epsilon_0_to_fp16, gamma = add_103_gamma_0_to_fp16, mean = add_97_mean_0_to_fp16, variance = add_97_variance_0_to_fp16, x = reshape_205_cast_fp16)[name = tensor("add_103_cast_fp16")]; tensor input_453_cast_fp16 = silu(x = add_103_cast_fp16)[name = tensor("input_453_cast_fp16")]; tensor var_10536 = const()[name = tensor("op_10536"), val = tensor([1, 1])]; tensor var_10538 = const()[name = tensor("op_10538"), val = tensor([1, 1])]; tensor hidden_states_279_pad_type_0 = const()[name = tensor("hidden_states_279_pad_type_0"), val = tensor("custom")]; tensor hidden_states_279_pad_0 = const()[name = tensor("hidden_states_279_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_3_resnets_0_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(799817152)))]; tensor up_blocks_3_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805346816)))]; tensor hidden_states_279_cast_fp16 = conv(bias = up_blocks_3_resnets_0_conv1_bias_to_fp16, dilations = var_10538, groups = var_10509, pad = hidden_states_279_pad_0, pad_type = hidden_states_279_pad_type_0, strides = var_10536, weight = up_blocks_3_resnets_0_conv1_weight_to_fp16, x = input_453_cast_fp16)[name = tensor("hidden_states_279_cast_fp16")]; tensor var_10544 = const()[name = tensor("op_10544"), val = tensor([1, 1])]; tensor var_10546 = const()[name = tensor("op_10546"), val = tensor([1, 1])]; tensor temb_39_pad_type_0 = const()[name = tensor("temb_39_pad_type_0"), val = tensor("custom")]; tensor temb_39_pad_0 = const()[name = tensor("temb_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805347520)))]; tensor up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(806166784)))]; tensor temb_39_cast_fp16 = conv(bias = up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_10546, groups = var_10509, pad = temb_39_pad_0, pad_type = temb_39_pad_type_0, strides = var_10544, weight = up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16, x = cast_12)[name = tensor("temb_39_cast_fp16")]; tensor input_457_cast_fp16 = add(x = hidden_states_279_cast_fp16, y = temb_39_cast_fp16)[name = tensor("input_457_cast_fp16")]; tensor reshape_208_shape_0 = const()[name = tensor("reshape_208_shape_0"), val = tensor([2, 32, 10, 48, 48])]; tensor reshape_208_cast_fp16 = reshape(shape = reshape_208_shape_0, x = input_457_cast_fp16)[name = tensor("reshape_208_cast_fp16")]; tensor reduce_mean_156_axes_0 = const()[name = tensor("reduce_mean_156_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_156_keep_dims_0 = const()[name = tensor("reduce_mean_156_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_156_cast_fp16 = reduce_mean(axes = reduce_mean_156_axes_0, keep_dims = reduce_mean_156_keep_dims_0, x = reshape_208_cast_fp16)[name = tensor("reduce_mean_156_cast_fp16")]; tensor sub_104_cast_fp16 = sub(x = reshape_208_cast_fp16, y = reduce_mean_156_cast_fp16)[name = tensor("sub_104_cast_fp16")]; tensor square_52_cast_fp16 = square(x = sub_104_cast_fp16)[name = tensor("square_52_cast_fp16")]; tensor reduce_mean_158_axes_0 = const()[name = tensor("reduce_mean_158_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_158_keep_dims_0 = const()[name = tensor("reduce_mean_158_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_158_cast_fp16 = reduce_mean(axes = reduce_mean_158_axes_0, keep_dims = reduce_mean_158_keep_dims_0, x = square_52_cast_fp16)[name = tensor("reduce_mean_158_cast_fp16")]; tensor add_104_y_0_to_fp16 = const()[name = tensor("add_104_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_104_cast_fp16 = add(x = reduce_mean_158_cast_fp16, y = add_104_y_0_to_fp16)[name = tensor("add_104_cast_fp16")]; tensor sqrt_52_cast_fp16 = sqrt(x = add_104_cast_fp16)[name = tensor("sqrt_52_cast_fp16")]; tensor real_div_52_cast_fp16 = real_div(x = sub_104_cast_fp16, y = sqrt_52_cast_fp16)[name = tensor("real_div_52_cast_fp16")]; tensor reshape_209_shape_0 = const()[name = tensor("reshape_209_shape_0"), val = tensor([2, 320, 48, 48])]; tensor reshape_209_cast_fp16 = reshape(shape = reshape_209_shape_0, x = real_div_52_cast_fp16)[name = tensor("reshape_209_cast_fp16")]; tensor add_105_gamma_0_to_fp16 = const()[name = tensor("add_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(806167488)))]; tensor add_105_beta_0_to_fp16 = const()[name = tensor("add_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(806168192)))]; tensor add_105_epsilon_0_to_fp16 = const()[name = tensor("add_105_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_105_cast_fp16 = batch_norm(beta = add_105_beta_0_to_fp16, epsilon = add_105_epsilon_0_to_fp16, gamma = add_105_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_209_cast_fp16)[name = tensor("add_105_cast_fp16")]; tensor input_461_cast_fp16 = silu(x = add_105_cast_fp16)[name = tensor("input_461_cast_fp16")]; tensor var_10556 = const()[name = tensor("op_10556"), val = tensor([1, 1])]; tensor var_10558 = const()[name = tensor("op_10558"), val = tensor([1, 1])]; tensor hidden_states_281_pad_type_0 = const()[name = tensor("hidden_states_281_pad_type_0"), val = tensor("custom")]; tensor hidden_states_281_pad_0 = const()[name = tensor("hidden_states_281_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_3_resnets_0_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(806168896)))]; tensor up_blocks_3_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808012160)))]; tensor hidden_states_281_cast_fp16 = conv(bias = up_blocks_3_resnets_0_conv2_bias_to_fp16, dilations = var_10558, groups = var_10509, pad = hidden_states_281_pad_0, pad_type = hidden_states_281_pad_type_0, strides = var_10556, weight = up_blocks_3_resnets_0_conv2_weight_to_fp16, x = input_461_cast_fp16)[name = tensor("hidden_states_281_cast_fp16")]; tensor var_10563 = const()[name = tensor("op_10563"), val = tensor([1, 1])]; tensor var_10565 = const()[name = tensor("op_10565"), val = tensor([1, 1])]; tensor x_23_pad_type_0 = const()[name = tensor("x_23_pad_type_0"), val = tensor("custom")]; tensor x_23_pad_0 = const()[name = tensor("x_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808012864)))]; tensor up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808627328)))]; tensor x_23_cast_fp16 = conv(bias = up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_10565, groups = var_10509, pad = x_23_pad_0, pad_type = x_23_pad_type_0, strides = var_10563, weight = up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16, x = input_449_cast_fp16)[name = tensor("x_23_cast_fp16")]; tensor hidden_states_283_cast_fp16 = add(x = x_23_cast_fp16, y = hidden_states_281_cast_fp16)[name = tensor("hidden_states_283_cast_fp16")]; tensor reshape_212_shape_0 = const()[name = tensor("reshape_212_shape_0"), val = tensor([2, 32, 10, 48, 48])]; tensor reshape_212_cast_fp16 = reshape(shape = reshape_212_shape_0, x = hidden_states_283_cast_fp16)[name = tensor("reshape_212_cast_fp16")]; tensor reduce_mean_159_axes_0 = const()[name = tensor("reduce_mean_159_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_159_keep_dims_0 = const()[name = tensor("reduce_mean_159_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_159_cast_fp16 = reduce_mean(axes = reduce_mean_159_axes_0, keep_dims = reduce_mean_159_keep_dims_0, x = reshape_212_cast_fp16)[name = tensor("reduce_mean_159_cast_fp16")]; tensor sub_106_cast_fp16 = sub(x = reshape_212_cast_fp16, y = reduce_mean_159_cast_fp16)[name = tensor("sub_106_cast_fp16")]; tensor square_53_cast_fp16 = square(x = sub_106_cast_fp16)[name = tensor("square_53_cast_fp16")]; tensor reduce_mean_161_axes_0 = const()[name = tensor("reduce_mean_161_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_161_keep_dims_0 = const()[name = tensor("reduce_mean_161_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_161_cast_fp16 = reduce_mean(axes = reduce_mean_161_axes_0, keep_dims = reduce_mean_161_keep_dims_0, x = square_53_cast_fp16)[name = tensor("reduce_mean_161_cast_fp16")]; tensor add_106_y_0_to_fp16 = const()[name = tensor("add_106_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_106_cast_fp16 = add(x = reduce_mean_161_cast_fp16, y = add_106_y_0_to_fp16)[name = tensor("add_106_cast_fp16")]; tensor sqrt_53_cast_fp16 = sqrt(x = add_106_cast_fp16)[name = tensor("sqrt_53_cast_fp16")]; tensor real_div_53_cast_fp16 = real_div(x = sub_106_cast_fp16, y = sqrt_53_cast_fp16)[name = tensor("real_div_53_cast_fp16")]; tensor reshape_213_shape_0 = const()[name = tensor("reshape_213_shape_0"), val = tensor([2, 320, 48, 48])]; tensor reshape_213_cast_fp16 = reshape(shape = reshape_213_shape_0, x = real_div_53_cast_fp16)[name = tensor("reshape_213_cast_fp16")]; tensor add_107_gamma_0_to_fp16 = const()[name = tensor("add_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808628032)))]; tensor add_107_beta_0_to_fp16 = const()[name = tensor("add_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808628736)))]; tensor add_107_epsilon_0_to_fp16 = const()[name = tensor("add_107_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_107_cast_fp16 = batch_norm(beta = add_107_beta_0_to_fp16, epsilon = add_107_epsilon_0_to_fp16, gamma = add_107_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_213_cast_fp16)[name = tensor("add_107_cast_fp16")]; tensor var_10585 = const()[name = tensor("op_10585"), val = tensor([1, 1])]; tensor var_10587 = const()[name = tensor("op_10587"), val = tensor([1, 1])]; tensor hidden_states_285_pad_type_0 = const()[name = tensor("hidden_states_285_pad_type_0"), val = tensor("custom")]; tensor hidden_states_285_pad_0 = const()[name = tensor("hidden_states_285_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808629440)))]; tensor up_blocks_3_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808834304)))]; tensor hidden_states_285_cast_fp16 = conv(bias = up_blocks_3_attentions_0_proj_in_bias_to_fp16, dilations = var_10587, groups = var_10509, pad = hidden_states_285_pad_0, pad_type = hidden_states_285_pad_type_0, strides = var_10585, weight = up_blocks_3_attentions_0_proj_in_weight_to_fp16, x = add_107_cast_fp16)[name = tensor("hidden_states_285_cast_fp16")]; tensor var_10592 = const()[name = tensor("op_10592"), val = tensor([2, 320, 1, 2304])]; tensor inputs_79_cast_fp16 = reshape(shape = var_10592, x = hidden_states_285_cast_fp16)[name = tensor("inputs_79_cast_fp16")]; tensor var_10602 = const()[name = tensor("op_10602"), val = tensor([1])]; tensor channels_mean_79_cast_fp16 = reduce_mean(axes = var_10602, keep_dims = var_10504, x = inputs_79_cast_fp16)[name = tensor("channels_mean_79_cast_fp16")]; tensor zero_mean_79_cast_fp16 = sub(x = inputs_79_cast_fp16, y = channels_mean_79_cast_fp16)[name = tensor("zero_mean_79_cast_fp16")]; tensor zero_mean_sq_79_cast_fp16 = mul(x = zero_mean_79_cast_fp16, y = zero_mean_79_cast_fp16)[name = tensor("zero_mean_sq_79_cast_fp16")]; tensor var_10606 = const()[name = tensor("op_10606"), val = tensor([1])]; tensor var_10607_cast_fp16 = reduce_mean(axes = var_10606, keep_dims = var_10504, x = zero_mean_sq_79_cast_fp16)[name = tensor("op_10607_cast_fp16")]; tensor var_10608_to_fp16 = const()[name = tensor("op_10608_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_10609_cast_fp16 = add(x = var_10607_cast_fp16, y = var_10608_to_fp16)[name = tensor("op_10609_cast_fp16")]; tensor denom_79_epsilon_0_to_fp16 = const()[name = tensor("denom_79_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_79_cast_fp16 = rsqrt(epsilon = denom_79_epsilon_0_to_fp16, x = var_10609_cast_fp16)[name = tensor("denom_79_cast_fp16")]; tensor out_79_cast_fp16 = mul(x = zero_mean_79_cast_fp16, y = denom_79_cast_fp16)[name = tensor("out_79_cast_fp16")]; tensor var_10613_to_fp16 = const()[name = tensor("op_10613_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808835008)))]; tensor var_10614_cast_fp16 = add(x = out_79_cast_fp16, y = var_10613_to_fp16)[name = tensor("op_10614_cast_fp16")]; tensor var_10616_to_fp16 = const()[name = tensor("op_10616_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808835712)))]; tensor hidden_states_287_cast_fp16 = mul(x = var_10614_cast_fp16, y = var_10616_to_fp16)[name = tensor("hidden_states_287_cast_fp16")]; tensor var_10623 = const()[name = tensor("op_10623"), val = tensor([1, 1])]; tensor var_10625 = const()[name = tensor("op_10625"), val = tensor([1, 1])]; tensor q_53_pad_type_0 = const()[name = tensor("q_53_pad_type_0"), val = tensor("custom")]; tensor q_53_pad_0 = const()[name = tensor("q_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808836416)))]; tensor q_53_cast_fp16 = conv(dilations = var_10625, groups = var_10509, pad = q_53_pad_0, pad_type = q_53_pad_type_0, strides = var_10623, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_287_cast_fp16)[name = tensor("q_53_cast_fp16")]; tensor var_10629 = const()[name = tensor("op_10629"), val = tensor([1, 1])]; tensor var_10631 = const()[name = tensor("op_10631"), val = tensor([1, 1])]; tensor k_105_pad_type_0 = const()[name = tensor("k_105_pad_type_0"), val = tensor("custom")]; tensor k_105_pad_0 = const()[name = tensor("k_105_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809041280)))]; tensor k_105_cast_fp16 = conv(dilations = var_10631, groups = var_10509, pad = k_105_pad_0, pad_type = k_105_pad_type_0, strides = var_10629, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_287_cast_fp16)[name = tensor("k_105_cast_fp16")]; tensor var_10635 = const()[name = tensor("op_10635"), val = tensor([1, 1])]; tensor var_10637 = const()[name = tensor("op_10637"), val = tensor([1, 1])]; tensor v_53_pad_type_0 = const()[name = tensor("v_53_pad_type_0"), val = tensor("custom")]; tensor v_53_pad_0 = const()[name = tensor("v_53_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809246144)))]; tensor v_53_cast_fp16 = conv(dilations = var_10637, groups = var_10509, pad = v_53_pad_0, pad_type = v_53_pad_type_0, strides = var_10635, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_287_cast_fp16)[name = tensor("v_53_cast_fp16")]; tensor var_10641_begin_0 = const()[name = tensor("op_10641_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10641_end_0 = const()[name = tensor("op_10641_end_0"), val = tensor([2, 64, 1, 2304])]; tensor var_10641_end_mask_0 = const()[name = tensor("op_10641_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10641_cast_fp16 = slice_by_index(begin = var_10641_begin_0, end = var_10641_end_0, end_mask = var_10641_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10641_cast_fp16")]; tensor var_10645_begin_0 = const()[name = tensor("op_10645_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_10645_end_0 = const()[name = tensor("op_10645_end_0"), val = tensor([2, 128, 1, 2304])]; tensor var_10645_end_mask_0 = const()[name = tensor("op_10645_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10645_cast_fp16 = slice_by_index(begin = var_10645_begin_0, end = var_10645_end_0, end_mask = var_10645_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10645_cast_fp16")]; tensor var_10649_begin_0 = const()[name = tensor("op_10649_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_10649_end_0 = const()[name = tensor("op_10649_end_0"), val = tensor([2, 192, 1, 2304])]; tensor var_10649_end_mask_0 = const()[name = tensor("op_10649_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10649_cast_fp16 = slice_by_index(begin = var_10649_begin_0, end = var_10649_end_0, end_mask = var_10649_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10649_cast_fp16")]; tensor var_10653_begin_0 = const()[name = tensor("op_10653_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_10653_end_0 = const()[name = tensor("op_10653_end_0"), val = tensor([2, 256, 1, 2304])]; tensor var_10653_end_mask_0 = const()[name = tensor("op_10653_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10653_cast_fp16 = slice_by_index(begin = var_10653_begin_0, end = var_10653_end_0, end_mask = var_10653_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10653_cast_fp16")]; tensor var_10657_begin_0 = const()[name = tensor("op_10657_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10657_end_0 = const()[name = tensor("op_10657_end_0"), val = tensor([2, 320, 1, 2304])]; tensor var_10657_end_mask_0 = const()[name = tensor("op_10657_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10657_cast_fp16 = slice_by_index(begin = var_10657_begin_0, end = var_10657_end_0, end_mask = var_10657_end_mask_0, x = q_53_cast_fp16)[name = tensor("op_10657_cast_fp16")]; tensor k_107_perm_0 = const()[name = tensor("k_107_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_10664_begin_0 = const()[name = tensor("op_10664_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10664_end_0 = const()[name = tensor("op_10664_end_0"), val = tensor([2, 2304, 1, 64])]; tensor var_10664_end_mask_0 = const()[name = tensor("op_10664_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_5 = transpose(perm = k_107_perm_0, x = k_105_cast_fp16)[name = tensor("transpose_5")]; tensor var_10664_cast_fp16 = slice_by_index(begin = var_10664_begin_0, end = var_10664_end_0, end_mask = var_10664_end_mask_0, x = transpose_5)[name = tensor("op_10664_cast_fp16")]; tensor var_10668_begin_0 = const()[name = tensor("op_10668_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_10668_end_0 = const()[name = tensor("op_10668_end_0"), val = tensor([2, 2304, 1, 128])]; tensor var_10668_end_mask_0 = const()[name = tensor("op_10668_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10668_cast_fp16 = slice_by_index(begin = var_10668_begin_0, end = var_10668_end_0, end_mask = var_10668_end_mask_0, x = transpose_5)[name = tensor("op_10668_cast_fp16")]; tensor var_10672_begin_0 = const()[name = tensor("op_10672_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_10672_end_0 = const()[name = tensor("op_10672_end_0"), val = tensor([2, 2304, 1, 192])]; tensor var_10672_end_mask_0 = const()[name = tensor("op_10672_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10672_cast_fp16 = slice_by_index(begin = var_10672_begin_0, end = var_10672_end_0, end_mask = var_10672_end_mask_0, x = transpose_5)[name = tensor("op_10672_cast_fp16")]; tensor var_10676_begin_0 = const()[name = tensor("op_10676_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_10676_end_0 = const()[name = tensor("op_10676_end_0"), val = tensor([2, 2304, 1, 256])]; tensor var_10676_end_mask_0 = const()[name = tensor("op_10676_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10676_cast_fp16 = slice_by_index(begin = var_10676_begin_0, end = var_10676_end_0, end_mask = var_10676_end_mask_0, x = transpose_5)[name = tensor("op_10676_cast_fp16")]; tensor var_10680_begin_0 = const()[name = tensor("op_10680_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_10680_end_0 = const()[name = tensor("op_10680_end_0"), val = tensor([2, 2304, 1, 320])]; tensor var_10680_end_mask_0 = const()[name = tensor("op_10680_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10680_cast_fp16 = slice_by_index(begin = var_10680_begin_0, end = var_10680_end_0, end_mask = var_10680_end_mask_0, x = transpose_5)[name = tensor("op_10680_cast_fp16")]; tensor var_10682_begin_0 = const()[name = tensor("op_10682_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10682_end_0 = const()[name = tensor("op_10682_end_0"), val = tensor([2, 64, 1, 2304])]; tensor var_10682_end_mask_0 = const()[name = tensor("op_10682_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10682_cast_fp16 = slice_by_index(begin = var_10682_begin_0, end = var_10682_end_0, end_mask = var_10682_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10682_cast_fp16")]; tensor var_10686_begin_0 = const()[name = tensor("op_10686_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_10686_end_0 = const()[name = tensor("op_10686_end_0"), val = tensor([2, 128, 1, 2304])]; tensor var_10686_end_mask_0 = const()[name = tensor("op_10686_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10686_cast_fp16 = slice_by_index(begin = var_10686_begin_0, end = var_10686_end_0, end_mask = var_10686_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10686_cast_fp16")]; tensor var_10690_begin_0 = const()[name = tensor("op_10690_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_10690_end_0 = const()[name = tensor("op_10690_end_0"), val = tensor([2, 192, 1, 2304])]; tensor var_10690_end_mask_0 = const()[name = tensor("op_10690_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10690_cast_fp16 = slice_by_index(begin = var_10690_begin_0, end = var_10690_end_0, end_mask = var_10690_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10690_cast_fp16")]; tensor var_10694_begin_0 = const()[name = tensor("op_10694_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_10694_end_0 = const()[name = tensor("op_10694_end_0"), val = tensor([2, 256, 1, 2304])]; tensor var_10694_end_mask_0 = const()[name = tensor("op_10694_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10694_cast_fp16 = slice_by_index(begin = var_10694_begin_0, end = var_10694_end_0, end_mask = var_10694_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10694_cast_fp16")]; tensor var_10698_begin_0 = const()[name = tensor("op_10698_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10698_end_0 = const()[name = tensor("op_10698_end_0"), val = tensor([2, 320, 1, 2304])]; tensor var_10698_end_mask_0 = const()[name = tensor("op_10698_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10698_cast_fp16 = slice_by_index(begin = var_10698_begin_0, end = var_10698_end_0, end_mask = var_10698_end_mask_0, x = v_53_cast_fp16)[name = tensor("op_10698_cast_fp16")]; tensor var_10702_equation_0 = const()[name = tensor("op_10702_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10702_cast_fp16 = einsum(equation = var_10702_equation_0, values = (var_10664_cast_fp16, var_10641_cast_fp16))[name = tensor("op_10702_cast_fp16")]; tensor var_10703_to_fp16 = const()[name = tensor("op_10703_to_fp16"), val = tensor(0x1p-3)]; tensor aw_721_cast_fp16 = mul(x = var_10702_cast_fp16, y = var_10703_to_fp16)[name = tensor("aw_721_cast_fp16")]; tensor var_10706_equation_0 = const()[name = tensor("op_10706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10706_cast_fp16 = einsum(equation = var_10706_equation_0, values = (var_10668_cast_fp16, var_10645_cast_fp16))[name = tensor("op_10706_cast_fp16")]; tensor var_10707_to_fp16 = const()[name = tensor("op_10707_to_fp16"), val = tensor(0x1p-3)]; tensor aw_723_cast_fp16 = mul(x = var_10706_cast_fp16, y = var_10707_to_fp16)[name = tensor("aw_723_cast_fp16")]; tensor var_10710_equation_0 = const()[name = tensor("op_10710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10710_cast_fp16 = einsum(equation = var_10710_equation_0, values = (var_10672_cast_fp16, var_10649_cast_fp16))[name = tensor("op_10710_cast_fp16")]; tensor var_10711_to_fp16 = const()[name = tensor("op_10711_to_fp16"), val = tensor(0x1p-3)]; tensor aw_725_cast_fp16 = mul(x = var_10710_cast_fp16, y = var_10711_to_fp16)[name = tensor("aw_725_cast_fp16")]; tensor var_10714_equation_0 = const()[name = tensor("op_10714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10714_cast_fp16 = einsum(equation = var_10714_equation_0, values = (var_10676_cast_fp16, var_10653_cast_fp16))[name = tensor("op_10714_cast_fp16")]; tensor var_10715_to_fp16 = const()[name = tensor("op_10715_to_fp16"), val = tensor(0x1p-3)]; tensor aw_727_cast_fp16 = mul(x = var_10714_cast_fp16, y = var_10715_to_fp16)[name = tensor("aw_727_cast_fp16")]; tensor var_10718_equation_0 = const()[name = tensor("op_10718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10718_cast_fp16 = einsum(equation = var_10718_equation_0, values = (var_10680_cast_fp16, var_10657_cast_fp16))[name = tensor("op_10718_cast_fp16")]; tensor var_10719_to_fp16 = const()[name = tensor("op_10719_to_fp16"), val = tensor(0x1p-3)]; tensor aw_729_cast_fp16 = mul(x = var_10718_cast_fp16, y = var_10719_to_fp16)[name = tensor("aw_729_cast_fp16")]; tensor var_10721_cast_fp16 = softmax(axis = var_10509, x = aw_721_cast_fp16)[name = tensor("op_10721_cast_fp16")]; tensor var_10722_cast_fp16 = softmax(axis = var_10509, x = aw_723_cast_fp16)[name = tensor("op_10722_cast_fp16")]; tensor var_10723_cast_fp16 = softmax(axis = var_10509, x = aw_725_cast_fp16)[name = tensor("op_10723_cast_fp16")]; tensor var_10724_cast_fp16 = softmax(axis = var_10509, x = aw_727_cast_fp16)[name = tensor("op_10724_cast_fp16")]; tensor var_10725_cast_fp16 = softmax(axis = var_10509, x = aw_729_cast_fp16)[name = tensor("op_10725_cast_fp16")]; tensor var_10727_equation_0 = const()[name = tensor("op_10727_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10727_cast_fp16 = einsum(equation = var_10727_equation_0, values = (var_10682_cast_fp16, var_10721_cast_fp16))[name = tensor("op_10727_cast_fp16")]; tensor var_10729_equation_0 = const()[name = tensor("op_10729_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10729_cast_fp16 = einsum(equation = var_10729_equation_0, values = (var_10686_cast_fp16, var_10722_cast_fp16))[name = tensor("op_10729_cast_fp16")]; tensor var_10731_equation_0 = const()[name = tensor("op_10731_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10731_cast_fp16 = einsum(equation = var_10731_equation_0, values = (var_10690_cast_fp16, var_10723_cast_fp16))[name = tensor("op_10731_cast_fp16")]; tensor var_10733_equation_0 = const()[name = tensor("op_10733_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10733_cast_fp16 = einsum(equation = var_10733_equation_0, values = (var_10694_cast_fp16, var_10724_cast_fp16))[name = tensor("op_10733_cast_fp16")]; tensor var_10735_equation_0 = const()[name = tensor("op_10735_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10735_cast_fp16 = einsum(equation = var_10735_equation_0, values = (var_10698_cast_fp16, var_10725_cast_fp16))[name = tensor("op_10735_cast_fp16")]; tensor input_465_interleave_0 = const()[name = tensor("input_465_interleave_0"), val = tensor(false)]; tensor input_465_cast_fp16 = concat(axis = var_10509, interleave = input_465_interleave_0, values = (var_10727_cast_fp16, var_10729_cast_fp16, var_10731_cast_fp16, var_10733_cast_fp16, var_10735_cast_fp16))[name = tensor("input_465_cast_fp16")]; tensor var_10741 = const()[name = tensor("op_10741"), val = tensor([1, 1])]; tensor var_10743 = const()[name = tensor("op_10743"), val = tensor([1, 1])]; tensor var_10745_pad_type_0 = const()[name = tensor("op_10745_pad_type_0"), val = tensor("custom")]; tensor var_10745_pad_0 = const()[name = tensor("op_10745_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809451008)))]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809655872)))]; tensor var_10745_cast_fp16 = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_10743, groups = var_10509, pad = var_10745_pad_0, pad_type = var_10745_pad_type_0, strides = var_10741, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_465_cast_fp16)[name = tensor("op_10745_cast_fp16")]; tensor inputs_81_cast_fp16 = add(x = var_10745_cast_fp16, y = inputs_79_cast_fp16)[name = tensor("inputs_81_cast_fp16")]; tensor var_10749 = const()[name = tensor("op_10749"), val = tensor([1])]; tensor channels_mean_81_cast_fp16 = reduce_mean(axes = var_10749, keep_dims = var_10504, x = inputs_81_cast_fp16)[name = tensor("channels_mean_81_cast_fp16")]; tensor zero_mean_81_cast_fp16 = sub(x = inputs_81_cast_fp16, y = channels_mean_81_cast_fp16)[name = tensor("zero_mean_81_cast_fp16")]; tensor zero_mean_sq_81_cast_fp16 = mul(x = zero_mean_81_cast_fp16, y = zero_mean_81_cast_fp16)[name = tensor("zero_mean_sq_81_cast_fp16")]; tensor var_10753 = const()[name = tensor("op_10753"), val = tensor([1])]; tensor var_10754_cast_fp16 = reduce_mean(axes = var_10753, keep_dims = var_10504, x = zero_mean_sq_81_cast_fp16)[name = tensor("op_10754_cast_fp16")]; tensor var_10755_to_fp16 = const()[name = tensor("op_10755_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_10756_cast_fp16 = add(x = var_10754_cast_fp16, y = var_10755_to_fp16)[name = tensor("op_10756_cast_fp16")]; tensor denom_81_epsilon_0_to_fp16 = const()[name = tensor("denom_81_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_81_cast_fp16 = rsqrt(epsilon = denom_81_epsilon_0_to_fp16, x = var_10756_cast_fp16)[name = tensor("denom_81_cast_fp16")]; tensor out_81_cast_fp16 = mul(x = zero_mean_81_cast_fp16, y = denom_81_cast_fp16)[name = tensor("out_81_cast_fp16")]; tensor var_10760_to_fp16 = const()[name = tensor("op_10760_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809656576)))]; tensor var_10761_cast_fp16 = add(x = out_81_cast_fp16, y = var_10760_to_fp16)[name = tensor("op_10761_cast_fp16")]; tensor var_10763_to_fp16 = const()[name = tensor("op_10763_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809657280)))]; tensor hidden_states_289_cast_fp16 = mul(x = var_10761_cast_fp16, y = var_10763_to_fp16)[name = tensor("hidden_states_289_cast_fp16")]; tensor var_10770 = const()[name = tensor("op_10770"), val = tensor([1, 1])]; tensor var_10772 = const()[name = tensor("op_10772"), val = tensor([1, 1])]; tensor q_55_pad_type_0 = const()[name = tensor("q_55_pad_type_0"), val = tensor("custom")]; tensor q_55_pad_0 = const()[name = tensor("q_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809657984)))]; tensor q_55_cast_fp16 = conv(dilations = var_10772, groups = var_10509, pad = q_55_pad_0, pad_type = q_55_pad_type_0, strides = var_10770, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_289_cast_fp16)[name = tensor("q_55_cast_fp16")]; tensor var_10776 = const()[name = tensor("op_10776"), val = tensor([1, 1])]; tensor var_10778 = const()[name = tensor("op_10778"), val = tensor([1, 1])]; tensor k_109_pad_type_0 = const()[name = tensor("k_109_pad_type_0"), val = tensor("custom")]; tensor k_109_pad_0 = const()[name = tensor("k_109_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(809862848)))]; tensor k_109_cast_fp16 = conv(dilations = var_10778, groups = var_10509, pad = k_109_pad_0, pad_type = k_109_pad_type_0, strides = var_10776, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_109_cast_fp16")]; tensor var_10782 = const()[name = tensor("op_10782"), val = tensor([1, 1])]; tensor var_10784 = const()[name = tensor("op_10784"), val = tensor([1, 1])]; tensor v_55_pad_type_0 = const()[name = tensor("v_55_pad_type_0"), val = tensor("custom")]; tensor v_55_pad_0 = const()[name = tensor("v_55_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(810518272)))]; tensor v_55_cast_fp16 = conv(dilations = var_10784, groups = var_10509, pad = v_55_pad_0, pad_type = v_55_pad_type_0, strides = var_10782, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_55_cast_fp16")]; tensor var_10788_begin_0 = const()[name = tensor("op_10788_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10788_end_0 = const()[name = tensor("op_10788_end_0"), val = tensor([2, 64, 1, 2304])]; tensor var_10788_end_mask_0 = const()[name = tensor("op_10788_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10788_cast_fp16 = slice_by_index(begin = var_10788_begin_0, end = var_10788_end_0, end_mask = var_10788_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_10788_cast_fp16")]; tensor var_10792_begin_0 = const()[name = tensor("op_10792_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_10792_end_0 = const()[name = tensor("op_10792_end_0"), val = tensor([2, 128, 1, 2304])]; tensor var_10792_end_mask_0 = const()[name = tensor("op_10792_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10792_cast_fp16 = slice_by_index(begin = var_10792_begin_0, end = var_10792_end_0, end_mask = var_10792_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_10792_cast_fp16")]; tensor var_10796_begin_0 = const()[name = tensor("op_10796_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_10796_end_0 = const()[name = tensor("op_10796_end_0"), val = tensor([2, 192, 1, 2304])]; tensor var_10796_end_mask_0 = const()[name = tensor("op_10796_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10796_cast_fp16 = slice_by_index(begin = var_10796_begin_0, end = var_10796_end_0, end_mask = var_10796_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_10796_cast_fp16")]; tensor var_10800_begin_0 = const()[name = tensor("op_10800_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_10800_end_0 = const()[name = tensor("op_10800_end_0"), val = tensor([2, 256, 1, 2304])]; tensor var_10800_end_mask_0 = const()[name = tensor("op_10800_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10800_cast_fp16 = slice_by_index(begin = var_10800_begin_0, end = var_10800_end_0, end_mask = var_10800_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_10800_cast_fp16")]; tensor var_10804_begin_0 = const()[name = tensor("op_10804_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10804_end_0 = const()[name = tensor("op_10804_end_0"), val = tensor([2, 320, 1, 2304])]; tensor var_10804_end_mask_0 = const()[name = tensor("op_10804_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10804_cast_fp16 = slice_by_index(begin = var_10804_begin_0, end = var_10804_end_0, end_mask = var_10804_end_mask_0, x = q_55_cast_fp16)[name = tensor("op_10804_cast_fp16")]; tensor k_111_perm_0 = const()[name = tensor("k_111_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_10811_begin_0 = const()[name = tensor("op_10811_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10811_end_0 = const()[name = tensor("op_10811_end_0"), val = tensor([2, 77, 1, 64])]; tensor var_10811_end_mask_0 = const()[name = tensor("op_10811_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_4 = transpose(perm = k_111_perm_0, x = k_109_cast_fp16)[name = tensor("transpose_4")]; tensor var_10811_cast_fp16 = slice_by_index(begin = var_10811_begin_0, end = var_10811_end_0, end_mask = var_10811_end_mask_0, x = transpose_4)[name = tensor("op_10811_cast_fp16")]; tensor var_10815_begin_0 = const()[name = tensor("op_10815_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_10815_end_0 = const()[name = tensor("op_10815_end_0"), val = tensor([2, 77, 1, 128])]; tensor var_10815_end_mask_0 = const()[name = tensor("op_10815_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10815_cast_fp16 = slice_by_index(begin = var_10815_begin_0, end = var_10815_end_0, end_mask = var_10815_end_mask_0, x = transpose_4)[name = tensor("op_10815_cast_fp16")]; tensor var_10819_begin_0 = const()[name = tensor("op_10819_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_10819_end_0 = const()[name = tensor("op_10819_end_0"), val = tensor([2, 77, 1, 192])]; tensor var_10819_end_mask_0 = const()[name = tensor("op_10819_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10819_cast_fp16 = slice_by_index(begin = var_10819_begin_0, end = var_10819_end_0, end_mask = var_10819_end_mask_0, x = transpose_4)[name = tensor("op_10819_cast_fp16")]; tensor var_10823_begin_0 = const()[name = tensor("op_10823_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_10823_end_0 = const()[name = tensor("op_10823_end_0"), val = tensor([2, 77, 1, 256])]; tensor var_10823_end_mask_0 = const()[name = tensor("op_10823_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10823_cast_fp16 = slice_by_index(begin = var_10823_begin_0, end = var_10823_end_0, end_mask = var_10823_end_mask_0, x = transpose_4)[name = tensor("op_10823_cast_fp16")]; tensor var_10827_begin_0 = const()[name = tensor("op_10827_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_10827_end_0 = const()[name = tensor("op_10827_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_10827_end_mask_0 = const()[name = tensor("op_10827_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10827_cast_fp16 = slice_by_index(begin = var_10827_begin_0, end = var_10827_end_0, end_mask = var_10827_end_mask_0, x = transpose_4)[name = tensor("op_10827_cast_fp16")]; tensor var_10829_begin_0 = const()[name = tensor("op_10829_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10829_end_0 = const()[name = tensor("op_10829_end_0"), val = tensor([2, 64, 1, 77])]; tensor var_10829_end_mask_0 = const()[name = tensor("op_10829_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10829_cast_fp16 = slice_by_index(begin = var_10829_begin_0, end = var_10829_end_0, end_mask = var_10829_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_10829_cast_fp16")]; tensor var_10833_begin_0 = const()[name = tensor("op_10833_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_10833_end_0 = const()[name = tensor("op_10833_end_0"), val = tensor([2, 128, 1, 77])]; tensor var_10833_end_mask_0 = const()[name = tensor("op_10833_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10833_cast_fp16 = slice_by_index(begin = var_10833_begin_0, end = var_10833_end_0, end_mask = var_10833_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_10833_cast_fp16")]; tensor var_10837_begin_0 = const()[name = tensor("op_10837_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_10837_end_0 = const()[name = tensor("op_10837_end_0"), val = tensor([2, 192, 1, 77])]; tensor var_10837_end_mask_0 = const()[name = tensor("op_10837_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10837_cast_fp16 = slice_by_index(begin = var_10837_begin_0, end = var_10837_end_0, end_mask = var_10837_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_10837_cast_fp16")]; tensor var_10841_begin_0 = const()[name = tensor("op_10841_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_10841_end_0 = const()[name = tensor("op_10841_end_0"), val = tensor([2, 256, 1, 77])]; tensor var_10841_end_mask_0 = const()[name = tensor("op_10841_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10841_cast_fp16 = slice_by_index(begin = var_10841_begin_0, end = var_10841_end_0, end_mask = var_10841_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_10841_cast_fp16")]; tensor var_10845_begin_0 = const()[name = tensor("op_10845_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10845_end_0 = const()[name = tensor("op_10845_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_10845_end_mask_0 = const()[name = tensor("op_10845_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10845_cast_fp16 = slice_by_index(begin = var_10845_begin_0, end = var_10845_end_0, end_mask = var_10845_end_mask_0, x = v_55_cast_fp16)[name = tensor("op_10845_cast_fp16")]; tensor var_10849_equation_0 = const()[name = tensor("op_10849_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10849_cast_fp16 = einsum(equation = var_10849_equation_0, values = (var_10811_cast_fp16, var_10788_cast_fp16))[name = tensor("op_10849_cast_fp16")]; tensor var_10850_to_fp16 = const()[name = tensor("op_10850_to_fp16"), val = tensor(0x1p-3)]; tensor aw_731_cast_fp16 = mul(x = var_10849_cast_fp16, y = var_10850_to_fp16)[name = tensor("aw_731_cast_fp16")]; tensor var_10853_equation_0 = const()[name = tensor("op_10853_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10853_cast_fp16 = einsum(equation = var_10853_equation_0, values = (var_10815_cast_fp16, var_10792_cast_fp16))[name = tensor("op_10853_cast_fp16")]; tensor var_10854_to_fp16 = const()[name = tensor("op_10854_to_fp16"), val = tensor(0x1p-3)]; tensor aw_733_cast_fp16 = mul(x = var_10853_cast_fp16, y = var_10854_to_fp16)[name = tensor("aw_733_cast_fp16")]; tensor var_10857_equation_0 = const()[name = tensor("op_10857_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10857_cast_fp16 = einsum(equation = var_10857_equation_0, values = (var_10819_cast_fp16, var_10796_cast_fp16))[name = tensor("op_10857_cast_fp16")]; tensor var_10858_to_fp16 = const()[name = tensor("op_10858_to_fp16"), val = tensor(0x1p-3)]; tensor aw_735_cast_fp16 = mul(x = var_10857_cast_fp16, y = var_10858_to_fp16)[name = tensor("aw_735_cast_fp16")]; tensor var_10861_equation_0 = const()[name = tensor("op_10861_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10861_cast_fp16 = einsum(equation = var_10861_equation_0, values = (var_10823_cast_fp16, var_10800_cast_fp16))[name = tensor("op_10861_cast_fp16")]; tensor var_10862_to_fp16 = const()[name = tensor("op_10862_to_fp16"), val = tensor(0x1p-3)]; tensor aw_737_cast_fp16 = mul(x = var_10861_cast_fp16, y = var_10862_to_fp16)[name = tensor("aw_737_cast_fp16")]; tensor var_10865_equation_0 = const()[name = tensor("op_10865_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10865_cast_fp16 = einsum(equation = var_10865_equation_0, values = (var_10827_cast_fp16, var_10804_cast_fp16))[name = tensor("op_10865_cast_fp16")]; tensor var_10866_to_fp16 = const()[name = tensor("op_10866_to_fp16"), val = tensor(0x1p-3)]; tensor aw_739_cast_fp16 = mul(x = var_10865_cast_fp16, y = var_10866_to_fp16)[name = tensor("aw_739_cast_fp16")]; tensor var_10868_cast_fp16 = softmax(axis = var_10509, x = aw_731_cast_fp16)[name = tensor("op_10868_cast_fp16")]; tensor var_10869_cast_fp16 = softmax(axis = var_10509, x = aw_733_cast_fp16)[name = tensor("op_10869_cast_fp16")]; tensor var_10870_cast_fp16 = softmax(axis = var_10509, x = aw_735_cast_fp16)[name = tensor("op_10870_cast_fp16")]; tensor var_10871_cast_fp16 = softmax(axis = var_10509, x = aw_737_cast_fp16)[name = tensor("op_10871_cast_fp16")]; tensor var_10872_cast_fp16 = softmax(axis = var_10509, x = aw_739_cast_fp16)[name = tensor("op_10872_cast_fp16")]; tensor var_10874_equation_0 = const()[name = tensor("op_10874_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10874_cast_fp16 = einsum(equation = var_10874_equation_0, values = (var_10829_cast_fp16, var_10868_cast_fp16))[name = tensor("op_10874_cast_fp16")]; tensor var_10876_equation_0 = const()[name = tensor("op_10876_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10876_cast_fp16 = einsum(equation = var_10876_equation_0, values = (var_10833_cast_fp16, var_10869_cast_fp16))[name = tensor("op_10876_cast_fp16")]; tensor var_10878_equation_0 = const()[name = tensor("op_10878_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10878_cast_fp16 = einsum(equation = var_10878_equation_0, values = (var_10837_cast_fp16, var_10870_cast_fp16))[name = tensor("op_10878_cast_fp16")]; tensor var_10880_equation_0 = const()[name = tensor("op_10880_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10880_cast_fp16 = einsum(equation = var_10880_equation_0, values = (var_10841_cast_fp16, var_10871_cast_fp16))[name = tensor("op_10880_cast_fp16")]; tensor var_10882_equation_0 = const()[name = tensor("op_10882_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10882_cast_fp16 = einsum(equation = var_10882_equation_0, values = (var_10845_cast_fp16, var_10872_cast_fp16))[name = tensor("op_10882_cast_fp16")]; tensor input_467_interleave_0 = const()[name = tensor("input_467_interleave_0"), val = tensor(false)]; tensor input_467_cast_fp16 = concat(axis = var_10509, interleave = input_467_interleave_0, values = (var_10874_cast_fp16, var_10876_cast_fp16, var_10878_cast_fp16, var_10880_cast_fp16, var_10882_cast_fp16))[name = tensor("input_467_cast_fp16")]; tensor var_10888 = const()[name = tensor("op_10888"), val = tensor([1, 1])]; tensor var_10890 = const()[name = tensor("op_10890"), val = tensor([1, 1])]; tensor var_10892_pad_type_0 = const()[name = tensor("op_10892_pad_type_0"), val = tensor("custom")]; tensor var_10892_pad_0 = const()[name = tensor("op_10892_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811173696)))]; tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811378560)))]; tensor var_10892_cast_fp16 = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_10890, groups = var_10509, pad = var_10892_pad_0, pad_type = var_10892_pad_type_0, strides = var_10888, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_467_cast_fp16)[name = tensor("op_10892_cast_fp16")]; tensor inputs_83_cast_fp16 = add(x = var_10892_cast_fp16, y = inputs_81_cast_fp16)[name = tensor("inputs_83_cast_fp16")]; tensor var_10896 = const()[name = tensor("op_10896"), val = tensor([1])]; tensor channels_mean_83_cast_fp16 = reduce_mean(axes = var_10896, keep_dims = var_10504, x = inputs_83_cast_fp16)[name = tensor("channels_mean_83_cast_fp16")]; tensor zero_mean_83_cast_fp16 = sub(x = inputs_83_cast_fp16, y = channels_mean_83_cast_fp16)[name = tensor("zero_mean_83_cast_fp16")]; tensor zero_mean_sq_83_cast_fp16 = mul(x = zero_mean_83_cast_fp16, y = zero_mean_83_cast_fp16)[name = tensor("zero_mean_sq_83_cast_fp16")]; tensor var_10900 = const()[name = tensor("op_10900"), val = tensor([1])]; tensor var_10901_cast_fp16 = reduce_mean(axes = var_10900, keep_dims = var_10504, x = zero_mean_sq_83_cast_fp16)[name = tensor("op_10901_cast_fp16")]; tensor var_10902_to_fp16 = const()[name = tensor("op_10902_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_10903_cast_fp16 = add(x = var_10901_cast_fp16, y = var_10902_to_fp16)[name = tensor("op_10903_cast_fp16")]; tensor denom_83_epsilon_0_to_fp16 = const()[name = tensor("denom_83_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_83_cast_fp16 = rsqrt(epsilon = denom_83_epsilon_0_to_fp16, x = var_10903_cast_fp16)[name = tensor("denom_83_cast_fp16")]; tensor out_83_cast_fp16 = mul(x = zero_mean_83_cast_fp16, y = denom_83_cast_fp16)[name = tensor("out_83_cast_fp16")]; tensor var_10907_to_fp16 = const()[name = tensor("op_10907_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811379264)))]; tensor var_10908_cast_fp16 = add(x = out_83_cast_fp16, y = var_10907_to_fp16)[name = tensor("op_10908_cast_fp16")]; tensor var_10910_to_fp16 = const()[name = tensor("op_10910_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811379968)))]; tensor input_469_cast_fp16 = mul(x = var_10908_cast_fp16, y = var_10910_to_fp16)[name = tensor("input_469_cast_fp16")]; tensor var_10918 = const()[name = tensor("op_10918"), val = tensor([1, 1])]; tensor var_10920 = const()[name = tensor("op_10920"), val = tensor([1, 1])]; tensor var_10922_pad_type_0 = const()[name = tensor("op_10922_pad_type_0"), val = tensor("custom")]; tensor var_10922_pad_0 = const()[name = tensor("op_10922_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811380672)))]; tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(813019136)))]; tensor var_10922_cast_fp16 = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_10920, groups = var_10509, pad = var_10922_pad_0, pad_type = var_10922_pad_type_0, strides = var_10918, weight = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_469_cast_fp16)[name = tensor("op_10922_cast_fp16")]; tensor var_10923_split_sizes_0 = const()[name = tensor("op_10923_split_sizes_0"), val = tensor([1280, 1280])]; tensor var_10923_axis_0 = const()[name = tensor("op_10923_axis_0"), val = tensor(1)]; tensor var_10923_cast_fp16_0, tensor var_10923_cast_fp16_1 = split(axis = var_10923_axis_0, split_sizes = var_10923_split_sizes_0, x = var_10922_cast_fp16)[name = tensor("op_10923_cast_fp16")]; tensor var_10925_mode_0 = const()[name = tensor("op_10925_mode_0"), val = tensor("EXACT")]; tensor var_10925_cast_fp16 = gelu(mode = var_10925_mode_0, x = var_10923_cast_fp16_1)[name = tensor("op_10925_cast_fp16")]; tensor input_471_cast_fp16 = mul(x = var_10923_cast_fp16_0, y = var_10925_cast_fp16)[name = tensor("input_471_cast_fp16")]; tensor var_10929 = const()[name = tensor("op_10929"), val = tensor([1, 1])]; tensor var_10931 = const()[name = tensor("op_10931"), val = tensor([1, 1])]; tensor var_10933_pad_type_0 = const()[name = tensor("op_10933_pad_type_0"), val = tensor("custom")]; tensor var_10933_pad_0 = const()[name = tensor("op_10933_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(813024320)))]; tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(813843584)))]; tensor var_10933_cast_fp16 = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_10931, groups = var_10509, pad = var_10933_pad_0, pad_type = var_10933_pad_type_0, strides = var_10929, weight = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_471_cast_fp16)[name = tensor("op_10933_cast_fp16")]; tensor hidden_states_293_cast_fp16 = add(x = var_10933_cast_fp16, y = inputs_83_cast_fp16)[name = tensor("hidden_states_293_cast_fp16")]; tensor var_10935 = const()[name = tensor("op_10935"), val = tensor([2, 320, 48, 48])]; tensor input_473_cast_fp16 = reshape(shape = var_10935, x = hidden_states_293_cast_fp16)[name = tensor("input_473_cast_fp16")]; tensor var_10939 = const()[name = tensor("op_10939"), val = tensor([1, 1])]; tensor var_10941 = const()[name = tensor("op_10941"), val = tensor([1, 1])]; tensor hidden_states_295_pad_type_0 = const()[name = tensor("hidden_states_295_pad_type_0"), val = tensor("custom")]; tensor hidden_states_295_pad_0 = const()[name = tensor("hidden_states_295_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_0_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(813844288)))]; tensor up_blocks_3_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814049152)))]; tensor hidden_states_295_cast_fp16 = conv(bias = up_blocks_3_attentions_0_proj_out_bias_to_fp16, dilations = var_10941, groups = var_10509, pad = hidden_states_295_pad_0, pad_type = hidden_states_295_pad_type_0, strides = var_10939, weight = up_blocks_3_attentions_0_proj_out_weight_to_fp16, x = input_473_cast_fp16)[name = tensor("hidden_states_295_cast_fp16")]; tensor hidden_states_297_cast_fp16 = add(x = hidden_states_295_cast_fp16, y = hidden_states_283_cast_fp16)[name = tensor("hidden_states_297_cast_fp16")]; tensor input_475_interleave_0 = const()[name = tensor("input_475_interleave_0"), val = tensor(false)]; tensor cast_11 = cast(dtype = cast_12_dtype_0, x = input_35_cast_fp16)[name = tensor("cast_11")]; tensor input_475_cast_fp16 = concat(axis = var_10509, interleave = input_475_interleave_0, values = (hidden_states_297_cast_fp16, cast_11))[name = tensor("input_475_cast_fp16")]; tensor reshape_216_shape_0 = const()[name = tensor("reshape_216_shape_0"), val = tensor([2, 32, 20, 48, 48])]; tensor reshape_216_cast_fp16 = reshape(shape = reshape_216_shape_0, x = input_475_cast_fp16)[name = tensor("reshape_216_cast_fp16")]; tensor reduce_mean_162_axes_0 = const()[name = tensor("reduce_mean_162_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_162_keep_dims_0 = const()[name = tensor("reduce_mean_162_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_162_cast_fp16 = reduce_mean(axes = reduce_mean_162_axes_0, keep_dims = reduce_mean_162_keep_dims_0, x = reshape_216_cast_fp16)[name = tensor("reduce_mean_162_cast_fp16")]; tensor sub_108_cast_fp16 = sub(x = reshape_216_cast_fp16, y = reduce_mean_162_cast_fp16)[name = tensor("sub_108_cast_fp16")]; tensor square_54_cast_fp16 = square(x = sub_108_cast_fp16)[name = tensor("square_54_cast_fp16")]; tensor reduce_mean_164_axes_0 = const()[name = tensor("reduce_mean_164_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_164_keep_dims_0 = const()[name = tensor("reduce_mean_164_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_164_cast_fp16 = reduce_mean(axes = reduce_mean_164_axes_0, keep_dims = reduce_mean_164_keep_dims_0, x = square_54_cast_fp16)[name = tensor("reduce_mean_164_cast_fp16")]; tensor add_108_y_0_to_fp16 = const()[name = tensor("add_108_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_108_cast_fp16 = add(x = reduce_mean_164_cast_fp16, y = add_108_y_0_to_fp16)[name = tensor("add_108_cast_fp16")]; tensor sqrt_54_cast_fp16 = sqrt(x = add_108_cast_fp16)[name = tensor("sqrt_54_cast_fp16")]; tensor real_div_54_cast_fp16 = real_div(x = sub_108_cast_fp16, y = sqrt_54_cast_fp16)[name = tensor("real_div_54_cast_fp16")]; tensor reshape_217_shape_0 = const()[name = tensor("reshape_217_shape_0"), val = tensor([2, 640, 48, 48])]; tensor reshape_217_cast_fp16 = reshape(shape = reshape_217_shape_0, x = real_div_54_cast_fp16)[name = tensor("reshape_217_cast_fp16")]; tensor add_109_gamma_0_to_fp16 = const()[name = tensor("add_109_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814049856)))]; tensor add_109_beta_0_to_fp16 = const()[name = tensor("add_109_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814051200)))]; tensor add_109_epsilon_0_to_fp16 = const()[name = tensor("add_109_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_109_cast_fp16 = batch_norm(beta = add_109_beta_0_to_fp16, epsilon = add_109_epsilon_0_to_fp16, gamma = add_109_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_217_cast_fp16)[name = tensor("add_109_cast_fp16")]; tensor input_479_cast_fp16 = silu(x = add_109_cast_fp16)[name = tensor("input_479_cast_fp16")]; tensor var_10959 = const()[name = tensor("op_10959"), val = tensor([1, 1])]; tensor var_10961 = const()[name = tensor("op_10961"), val = tensor([1, 1])]; tensor hidden_states_299_pad_type_0 = const()[name = tensor("hidden_states_299_pad_type_0"), val = tensor("custom")]; tensor hidden_states_299_pad_0 = const()[name = tensor("hidden_states_299_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_3_resnets_1_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814052544)))]; tensor up_blocks_3_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(817739008)))]; tensor hidden_states_299_cast_fp16 = conv(bias = up_blocks_3_resnets_1_conv1_bias_to_fp16, dilations = var_10961, groups = var_10509, pad = hidden_states_299_pad_0, pad_type = hidden_states_299_pad_type_0, strides = var_10959, weight = up_blocks_3_resnets_1_conv1_weight_to_fp16, x = input_479_cast_fp16)[name = tensor("hidden_states_299_cast_fp16")]; tensor var_10967 = const()[name = tensor("op_10967"), val = tensor([1, 1])]; tensor var_10969 = const()[name = tensor("op_10969"), val = tensor([1, 1])]; tensor temb_41_pad_type_0 = const()[name = tensor("temb_41_pad_type_0"), val = tensor("custom")]; tensor temb_41_pad_0 = const()[name = tensor("temb_41_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(817739712)))]; tensor up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(818558976)))]; tensor temb_41_cast_fp16 = conv(bias = up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_10969, groups = var_10509, pad = temb_41_pad_0, pad_type = temb_41_pad_type_0, strides = var_10967, weight = up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16, x = cast_12)[name = tensor("temb_41_cast_fp16")]; tensor input_483_cast_fp16 = add(x = hidden_states_299_cast_fp16, y = temb_41_cast_fp16)[name = tensor("input_483_cast_fp16")]; tensor reshape_220_shape_0 = const()[name = tensor("reshape_220_shape_0"), val = tensor([2, 32, 10, 48, 48])]; tensor reshape_220_cast_fp16 = reshape(shape = reshape_220_shape_0, x = input_483_cast_fp16)[name = tensor("reshape_220_cast_fp16")]; tensor reduce_mean_165_axes_0 = const()[name = tensor("reduce_mean_165_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_165_keep_dims_0 = const()[name = tensor("reduce_mean_165_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_165_cast_fp16 = reduce_mean(axes = reduce_mean_165_axes_0, keep_dims = reduce_mean_165_keep_dims_0, x = reshape_220_cast_fp16)[name = tensor("reduce_mean_165_cast_fp16")]; tensor sub_110_cast_fp16 = sub(x = reshape_220_cast_fp16, y = reduce_mean_165_cast_fp16)[name = tensor("sub_110_cast_fp16")]; tensor square_55_cast_fp16 = square(x = sub_110_cast_fp16)[name = tensor("square_55_cast_fp16")]; tensor reduce_mean_167_axes_0 = const()[name = tensor("reduce_mean_167_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_167_keep_dims_0 = const()[name = tensor("reduce_mean_167_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_167_cast_fp16 = reduce_mean(axes = reduce_mean_167_axes_0, keep_dims = reduce_mean_167_keep_dims_0, x = square_55_cast_fp16)[name = tensor("reduce_mean_167_cast_fp16")]; tensor add_110_y_0_to_fp16 = const()[name = tensor("add_110_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_110_cast_fp16 = add(x = reduce_mean_167_cast_fp16, y = add_110_y_0_to_fp16)[name = tensor("add_110_cast_fp16")]; tensor sqrt_55_cast_fp16 = sqrt(x = add_110_cast_fp16)[name = tensor("sqrt_55_cast_fp16")]; tensor real_div_55_cast_fp16 = real_div(x = sub_110_cast_fp16, y = sqrt_55_cast_fp16)[name = tensor("real_div_55_cast_fp16")]; tensor reshape_221_shape_0 = const()[name = tensor("reshape_221_shape_0"), val = tensor([2, 320, 48, 48])]; tensor reshape_221_cast_fp16 = reshape(shape = reshape_221_shape_0, x = real_div_55_cast_fp16)[name = tensor("reshape_221_cast_fp16")]; tensor add_111_gamma_0_to_fp16 = const()[name = tensor("add_111_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(818559680)))]; tensor add_111_beta_0_to_fp16 = const()[name = tensor("add_111_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(818560384)))]; tensor add_111_epsilon_0_to_fp16 = const()[name = tensor("add_111_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_111_cast_fp16 = batch_norm(beta = add_111_beta_0_to_fp16, epsilon = add_111_epsilon_0_to_fp16, gamma = add_111_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_221_cast_fp16)[name = tensor("add_111_cast_fp16")]; tensor input_487_cast_fp16 = silu(x = add_111_cast_fp16)[name = tensor("input_487_cast_fp16")]; tensor var_10979 = const()[name = tensor("op_10979"), val = tensor([1, 1])]; tensor var_10981 = const()[name = tensor("op_10981"), val = tensor([1, 1])]; tensor hidden_states_301_pad_type_0 = const()[name = tensor("hidden_states_301_pad_type_0"), val = tensor("custom")]; tensor hidden_states_301_pad_0 = const()[name = tensor("hidden_states_301_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_3_resnets_1_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(818561088)))]; tensor up_blocks_3_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(820404352)))]; tensor hidden_states_301_cast_fp16 = conv(bias = up_blocks_3_resnets_1_conv2_bias_to_fp16, dilations = var_10981, groups = var_10509, pad = hidden_states_301_pad_0, pad_type = hidden_states_301_pad_type_0, strides = var_10979, weight = up_blocks_3_resnets_1_conv2_weight_to_fp16, x = input_487_cast_fp16)[name = tensor("hidden_states_301_cast_fp16")]; tensor var_10986 = const()[name = tensor("op_10986"), val = tensor([1, 1])]; tensor var_10988 = const()[name = tensor("op_10988"), val = tensor([1, 1])]; tensor x_25_pad_type_0 = const()[name = tensor("x_25_pad_type_0"), val = tensor("custom")]; tensor x_25_pad_0 = const()[name = tensor("x_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(820405056)))]; tensor up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(820814720)))]; tensor x_25_cast_fp16 = conv(bias = up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_10988, groups = var_10509, pad = x_25_pad_0, pad_type = x_25_pad_type_0, strides = var_10986, weight = up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16, x = input_475_cast_fp16)[name = tensor("x_25_cast_fp16")]; tensor hidden_states_303_cast_fp16 = add(x = x_25_cast_fp16, y = hidden_states_301_cast_fp16)[name = tensor("hidden_states_303_cast_fp16")]; tensor reshape_224_shape_0 = const()[name = tensor("reshape_224_shape_0"), val = tensor([2, 32, 10, 48, 48])]; tensor reshape_224_cast_fp16 = reshape(shape = reshape_224_shape_0, x = hidden_states_303_cast_fp16)[name = tensor("reshape_224_cast_fp16")]; tensor reduce_mean_168_axes_0 = const()[name = tensor("reduce_mean_168_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_168_keep_dims_0 = const()[name = tensor("reduce_mean_168_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_168_cast_fp16 = reduce_mean(axes = reduce_mean_168_axes_0, keep_dims = reduce_mean_168_keep_dims_0, x = reshape_224_cast_fp16)[name = tensor("reduce_mean_168_cast_fp16")]; tensor sub_112_cast_fp16 = sub(x = reshape_224_cast_fp16, y = reduce_mean_168_cast_fp16)[name = tensor("sub_112_cast_fp16")]; tensor square_56_cast_fp16 = square(x = sub_112_cast_fp16)[name = tensor("square_56_cast_fp16")]; tensor reduce_mean_170_axes_0 = const()[name = tensor("reduce_mean_170_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_170_keep_dims_0 = const()[name = tensor("reduce_mean_170_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_170_cast_fp16 = reduce_mean(axes = reduce_mean_170_axes_0, keep_dims = reduce_mean_170_keep_dims_0, x = square_56_cast_fp16)[name = tensor("reduce_mean_170_cast_fp16")]; tensor add_112_y_0_to_fp16 = const()[name = tensor("add_112_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_112_cast_fp16 = add(x = reduce_mean_170_cast_fp16, y = add_112_y_0_to_fp16)[name = tensor("add_112_cast_fp16")]; tensor sqrt_56_cast_fp16 = sqrt(x = add_112_cast_fp16)[name = tensor("sqrt_56_cast_fp16")]; tensor real_div_56_cast_fp16 = real_div(x = sub_112_cast_fp16, y = sqrt_56_cast_fp16)[name = tensor("real_div_56_cast_fp16")]; tensor reshape_225_shape_0 = const()[name = tensor("reshape_225_shape_0"), val = tensor([2, 320, 48, 48])]; tensor reshape_225_cast_fp16 = reshape(shape = reshape_225_shape_0, x = real_div_56_cast_fp16)[name = tensor("reshape_225_cast_fp16")]; tensor add_113_gamma_0_to_fp16 = const()[name = tensor("add_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(820815424)))]; tensor add_113_beta_0_to_fp16 = const()[name = tensor("add_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(820816128)))]; tensor add_113_epsilon_0_to_fp16 = const()[name = tensor("add_113_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_113_cast_fp16 = batch_norm(beta = add_113_beta_0_to_fp16, epsilon = add_113_epsilon_0_to_fp16, gamma = add_113_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_225_cast_fp16)[name = tensor("add_113_cast_fp16")]; tensor var_11008 = const()[name = tensor("op_11008"), val = tensor([1, 1])]; tensor var_11010 = const()[name = tensor("op_11010"), val = tensor([1, 1])]; tensor hidden_states_305_pad_type_0 = const()[name = tensor("hidden_states_305_pad_type_0"), val = tensor("custom")]; tensor hidden_states_305_pad_0 = const()[name = tensor("hidden_states_305_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(820816832)))]; tensor up_blocks_3_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(821021696)))]; tensor hidden_states_305_cast_fp16 = conv(bias = up_blocks_3_attentions_1_proj_in_bias_to_fp16, dilations = var_11010, groups = var_10509, pad = hidden_states_305_pad_0, pad_type = hidden_states_305_pad_type_0, strides = var_11008, weight = up_blocks_3_attentions_1_proj_in_weight_to_fp16, x = add_113_cast_fp16)[name = tensor("hidden_states_305_cast_fp16")]; tensor var_11015 = const()[name = tensor("op_11015"), val = tensor([2, 320, 1, 2304])]; tensor inputs_85_cast_fp16 = reshape(shape = var_11015, x = hidden_states_305_cast_fp16)[name = tensor("inputs_85_cast_fp16")]; tensor var_11025 = const()[name = tensor("op_11025"), val = tensor([1])]; tensor channels_mean_85_cast_fp16 = reduce_mean(axes = var_11025, keep_dims = var_10504, x = inputs_85_cast_fp16)[name = tensor("channels_mean_85_cast_fp16")]; tensor zero_mean_85_cast_fp16 = sub(x = inputs_85_cast_fp16, y = channels_mean_85_cast_fp16)[name = tensor("zero_mean_85_cast_fp16")]; tensor zero_mean_sq_85_cast_fp16 = mul(x = zero_mean_85_cast_fp16, y = zero_mean_85_cast_fp16)[name = tensor("zero_mean_sq_85_cast_fp16")]; tensor var_11029 = const()[name = tensor("op_11029"), val = tensor([1])]; tensor var_11030_cast_fp16 = reduce_mean(axes = var_11029, keep_dims = var_10504, x = zero_mean_sq_85_cast_fp16)[name = tensor("op_11030_cast_fp16")]; tensor var_11031_to_fp16 = const()[name = tensor("op_11031_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_11032_cast_fp16 = add(x = var_11030_cast_fp16, y = var_11031_to_fp16)[name = tensor("op_11032_cast_fp16")]; tensor denom_85_epsilon_0_to_fp16 = const()[name = tensor("denom_85_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_85_cast_fp16 = rsqrt(epsilon = denom_85_epsilon_0_to_fp16, x = var_11032_cast_fp16)[name = tensor("denom_85_cast_fp16")]; tensor out_85_cast_fp16 = mul(x = zero_mean_85_cast_fp16, y = denom_85_cast_fp16)[name = tensor("out_85_cast_fp16")]; tensor var_11036_to_fp16 = const()[name = tensor("op_11036_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(821022400)))]; tensor var_11037_cast_fp16 = add(x = out_85_cast_fp16, y = var_11036_to_fp16)[name = tensor("op_11037_cast_fp16")]; tensor var_11039_to_fp16 = const()[name = tensor("op_11039_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(821023104)))]; tensor hidden_states_307_cast_fp16 = mul(x = var_11037_cast_fp16, y = var_11039_to_fp16)[name = tensor("hidden_states_307_cast_fp16")]; tensor var_11046 = const()[name = tensor("op_11046"), val = tensor([1, 1])]; tensor var_11048 = const()[name = tensor("op_11048"), val = tensor([1, 1])]; tensor q_57_pad_type_0 = const()[name = tensor("q_57_pad_type_0"), val = tensor("custom")]; tensor q_57_pad_0 = const()[name = tensor("q_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(821023808)))]; tensor q_57_cast_fp16 = conv(dilations = var_11048, groups = var_10509, pad = q_57_pad_0, pad_type = q_57_pad_type_0, strides = var_11046, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_307_cast_fp16)[name = tensor("q_57_cast_fp16")]; tensor var_11052 = const()[name = tensor("op_11052"), val = tensor([1, 1])]; tensor var_11054 = const()[name = tensor("op_11054"), val = tensor([1, 1])]; tensor k_113_pad_type_0 = const()[name = tensor("k_113_pad_type_0"), val = tensor("custom")]; tensor k_113_pad_0 = const()[name = tensor("k_113_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(821228672)))]; tensor k_113_cast_fp16 = conv(dilations = var_11054, groups = var_10509, pad = k_113_pad_0, pad_type = k_113_pad_type_0, strides = var_11052, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_307_cast_fp16)[name = tensor("k_113_cast_fp16")]; tensor var_11058 = const()[name = tensor("op_11058"), val = tensor([1, 1])]; tensor var_11060 = const()[name = tensor("op_11060"), val = tensor([1, 1])]; tensor v_57_pad_type_0 = const()[name = tensor("v_57_pad_type_0"), val = tensor("custom")]; tensor v_57_pad_0 = const()[name = tensor("v_57_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(821433536)))]; tensor v_57_cast_fp16 = conv(dilations = var_11060, groups = var_10509, pad = v_57_pad_0, pad_type = v_57_pad_type_0, strides = var_11058, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_307_cast_fp16)[name = tensor("v_57_cast_fp16")]; tensor var_11064_begin_0 = const()[name = tensor("op_11064_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11064_end_0 = const()[name = tensor("op_11064_end_0"), val = tensor([2, 64, 1, 2304])]; tensor var_11064_end_mask_0 = const()[name = tensor("op_11064_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11064_cast_fp16 = slice_by_index(begin = var_11064_begin_0, end = var_11064_end_0, end_mask = var_11064_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11064_cast_fp16")]; tensor var_11068_begin_0 = const()[name = tensor("op_11068_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_11068_end_0 = const()[name = tensor("op_11068_end_0"), val = tensor([2, 128, 1, 2304])]; tensor var_11068_end_mask_0 = const()[name = tensor("op_11068_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11068_cast_fp16 = slice_by_index(begin = var_11068_begin_0, end = var_11068_end_0, end_mask = var_11068_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11068_cast_fp16")]; tensor var_11072_begin_0 = const()[name = tensor("op_11072_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_11072_end_0 = const()[name = tensor("op_11072_end_0"), val = tensor([2, 192, 1, 2304])]; tensor var_11072_end_mask_0 = const()[name = tensor("op_11072_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11072_cast_fp16 = slice_by_index(begin = var_11072_begin_0, end = var_11072_end_0, end_mask = var_11072_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11072_cast_fp16")]; tensor var_11076_begin_0 = const()[name = tensor("op_11076_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_11076_end_0 = const()[name = tensor("op_11076_end_0"), val = tensor([2, 256, 1, 2304])]; tensor var_11076_end_mask_0 = const()[name = tensor("op_11076_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11076_cast_fp16 = slice_by_index(begin = var_11076_begin_0, end = var_11076_end_0, end_mask = var_11076_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11076_cast_fp16")]; tensor var_11080_begin_0 = const()[name = tensor("op_11080_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_11080_end_0 = const()[name = tensor("op_11080_end_0"), val = tensor([2, 320, 1, 2304])]; tensor var_11080_end_mask_0 = const()[name = tensor("op_11080_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11080_cast_fp16 = slice_by_index(begin = var_11080_begin_0, end = var_11080_end_0, end_mask = var_11080_end_mask_0, x = q_57_cast_fp16)[name = tensor("op_11080_cast_fp16")]; tensor k_115_perm_0 = const()[name = tensor("k_115_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_11087_begin_0 = const()[name = tensor("op_11087_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11087_end_0 = const()[name = tensor("op_11087_end_0"), val = tensor([2, 2304, 1, 64])]; tensor var_11087_end_mask_0 = const()[name = tensor("op_11087_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_3 = transpose(perm = k_115_perm_0, x = k_113_cast_fp16)[name = tensor("transpose_3")]; tensor var_11087_cast_fp16 = slice_by_index(begin = var_11087_begin_0, end = var_11087_end_0, end_mask = var_11087_end_mask_0, x = transpose_3)[name = tensor("op_11087_cast_fp16")]; tensor var_11091_begin_0 = const()[name = tensor("op_11091_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_11091_end_0 = const()[name = tensor("op_11091_end_0"), val = tensor([2, 2304, 1, 128])]; tensor var_11091_end_mask_0 = const()[name = tensor("op_11091_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11091_cast_fp16 = slice_by_index(begin = var_11091_begin_0, end = var_11091_end_0, end_mask = var_11091_end_mask_0, x = transpose_3)[name = tensor("op_11091_cast_fp16")]; tensor var_11095_begin_0 = const()[name = tensor("op_11095_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_11095_end_0 = const()[name = tensor("op_11095_end_0"), val = tensor([2, 2304, 1, 192])]; tensor var_11095_end_mask_0 = const()[name = tensor("op_11095_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11095_cast_fp16 = slice_by_index(begin = var_11095_begin_0, end = var_11095_end_0, end_mask = var_11095_end_mask_0, x = transpose_3)[name = tensor("op_11095_cast_fp16")]; tensor var_11099_begin_0 = const()[name = tensor("op_11099_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_11099_end_0 = const()[name = tensor("op_11099_end_0"), val = tensor([2, 2304, 1, 256])]; tensor var_11099_end_mask_0 = const()[name = tensor("op_11099_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11099_cast_fp16 = slice_by_index(begin = var_11099_begin_0, end = var_11099_end_0, end_mask = var_11099_end_mask_0, x = transpose_3)[name = tensor("op_11099_cast_fp16")]; tensor var_11103_begin_0 = const()[name = tensor("op_11103_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_11103_end_0 = const()[name = tensor("op_11103_end_0"), val = tensor([2, 2304, 1, 320])]; tensor var_11103_end_mask_0 = const()[name = tensor("op_11103_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11103_cast_fp16 = slice_by_index(begin = var_11103_begin_0, end = var_11103_end_0, end_mask = var_11103_end_mask_0, x = transpose_3)[name = tensor("op_11103_cast_fp16")]; tensor var_11105_begin_0 = const()[name = tensor("op_11105_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11105_end_0 = const()[name = tensor("op_11105_end_0"), val = tensor([2, 64, 1, 2304])]; tensor var_11105_end_mask_0 = const()[name = tensor("op_11105_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11105_cast_fp16 = slice_by_index(begin = var_11105_begin_0, end = var_11105_end_0, end_mask = var_11105_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_11105_cast_fp16")]; tensor var_11109_begin_0 = const()[name = tensor("op_11109_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_11109_end_0 = const()[name = tensor("op_11109_end_0"), val = tensor([2, 128, 1, 2304])]; tensor var_11109_end_mask_0 = const()[name = tensor("op_11109_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11109_cast_fp16 = slice_by_index(begin = var_11109_begin_0, end = var_11109_end_0, end_mask = var_11109_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_11109_cast_fp16")]; tensor var_11113_begin_0 = const()[name = tensor("op_11113_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_11113_end_0 = const()[name = tensor("op_11113_end_0"), val = tensor([2, 192, 1, 2304])]; tensor var_11113_end_mask_0 = const()[name = tensor("op_11113_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11113_cast_fp16 = slice_by_index(begin = var_11113_begin_0, end = var_11113_end_0, end_mask = var_11113_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_11113_cast_fp16")]; tensor var_11117_begin_0 = const()[name = tensor("op_11117_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_11117_end_0 = const()[name = tensor("op_11117_end_0"), val = tensor([2, 256, 1, 2304])]; tensor var_11117_end_mask_0 = const()[name = tensor("op_11117_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11117_cast_fp16 = slice_by_index(begin = var_11117_begin_0, end = var_11117_end_0, end_mask = var_11117_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_11117_cast_fp16")]; tensor var_11121_begin_0 = const()[name = tensor("op_11121_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_11121_end_0 = const()[name = tensor("op_11121_end_0"), val = tensor([2, 320, 1, 2304])]; tensor var_11121_end_mask_0 = const()[name = tensor("op_11121_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11121_cast_fp16 = slice_by_index(begin = var_11121_begin_0, end = var_11121_end_0, end_mask = var_11121_end_mask_0, x = v_57_cast_fp16)[name = tensor("op_11121_cast_fp16")]; tensor var_11125_equation_0 = const()[name = tensor("op_11125_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11125_cast_fp16 = einsum(equation = var_11125_equation_0, values = (var_11087_cast_fp16, var_11064_cast_fp16))[name = tensor("op_11125_cast_fp16")]; tensor var_11126_to_fp16 = const()[name = tensor("op_11126_to_fp16"), val = tensor(0x1p-3)]; tensor aw_741_cast_fp16 = mul(x = var_11125_cast_fp16, y = var_11126_to_fp16)[name = tensor("aw_741_cast_fp16")]; tensor var_11129_equation_0 = const()[name = tensor("op_11129_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11129_cast_fp16 = einsum(equation = var_11129_equation_0, values = (var_11091_cast_fp16, var_11068_cast_fp16))[name = tensor("op_11129_cast_fp16")]; tensor var_11130_to_fp16 = const()[name = tensor("op_11130_to_fp16"), val = tensor(0x1p-3)]; tensor aw_743_cast_fp16 = mul(x = var_11129_cast_fp16, y = var_11130_to_fp16)[name = tensor("aw_743_cast_fp16")]; tensor var_11133_equation_0 = const()[name = tensor("op_11133_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11133_cast_fp16 = einsum(equation = var_11133_equation_0, values = (var_11095_cast_fp16, var_11072_cast_fp16))[name = tensor("op_11133_cast_fp16")]; tensor var_11134_to_fp16 = const()[name = tensor("op_11134_to_fp16"), val = tensor(0x1p-3)]; tensor aw_745_cast_fp16 = mul(x = var_11133_cast_fp16, y = var_11134_to_fp16)[name = tensor("aw_745_cast_fp16")]; tensor var_11137_equation_0 = const()[name = tensor("op_11137_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11137_cast_fp16 = einsum(equation = var_11137_equation_0, values = (var_11099_cast_fp16, var_11076_cast_fp16))[name = tensor("op_11137_cast_fp16")]; tensor var_11138_to_fp16 = const()[name = tensor("op_11138_to_fp16"), val = tensor(0x1p-3)]; tensor aw_747_cast_fp16 = mul(x = var_11137_cast_fp16, y = var_11138_to_fp16)[name = tensor("aw_747_cast_fp16")]; tensor var_11141_equation_0 = const()[name = tensor("op_11141_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11141_cast_fp16 = einsum(equation = var_11141_equation_0, values = (var_11103_cast_fp16, var_11080_cast_fp16))[name = tensor("op_11141_cast_fp16")]; tensor var_11142_to_fp16 = const()[name = tensor("op_11142_to_fp16"), val = tensor(0x1p-3)]; tensor aw_749_cast_fp16 = mul(x = var_11141_cast_fp16, y = var_11142_to_fp16)[name = tensor("aw_749_cast_fp16")]; tensor var_11144_cast_fp16 = softmax(axis = var_10509, x = aw_741_cast_fp16)[name = tensor("op_11144_cast_fp16")]; tensor var_11145_cast_fp16 = softmax(axis = var_10509, x = aw_743_cast_fp16)[name = tensor("op_11145_cast_fp16")]; tensor var_11146_cast_fp16 = softmax(axis = var_10509, x = aw_745_cast_fp16)[name = tensor("op_11146_cast_fp16")]; tensor var_11147_cast_fp16 = softmax(axis = var_10509, x = aw_747_cast_fp16)[name = tensor("op_11147_cast_fp16")]; tensor var_11148_cast_fp16 = softmax(axis = var_10509, x = aw_749_cast_fp16)[name = tensor("op_11148_cast_fp16")]; tensor var_11150_equation_0 = const()[name = tensor("op_11150_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11150_cast_fp16 = einsum(equation = var_11150_equation_0, values = (var_11105_cast_fp16, var_11144_cast_fp16))[name = tensor("op_11150_cast_fp16")]; tensor var_11152_equation_0 = const()[name = tensor("op_11152_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11152_cast_fp16 = einsum(equation = var_11152_equation_0, values = (var_11109_cast_fp16, var_11145_cast_fp16))[name = tensor("op_11152_cast_fp16")]; tensor var_11154_equation_0 = const()[name = tensor("op_11154_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11154_cast_fp16 = einsum(equation = var_11154_equation_0, values = (var_11113_cast_fp16, var_11146_cast_fp16))[name = tensor("op_11154_cast_fp16")]; tensor var_11156_equation_0 = const()[name = tensor("op_11156_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11156_cast_fp16 = einsum(equation = var_11156_equation_0, values = (var_11117_cast_fp16, var_11147_cast_fp16))[name = tensor("op_11156_cast_fp16")]; tensor var_11158_equation_0 = const()[name = tensor("op_11158_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11158_cast_fp16 = einsum(equation = var_11158_equation_0, values = (var_11121_cast_fp16, var_11148_cast_fp16))[name = tensor("op_11158_cast_fp16")]; tensor input_491_interleave_0 = const()[name = tensor("input_491_interleave_0"), val = tensor(false)]; tensor input_491_cast_fp16 = concat(axis = var_10509, interleave = input_491_interleave_0, values = (var_11150_cast_fp16, var_11152_cast_fp16, var_11154_cast_fp16, var_11156_cast_fp16, var_11158_cast_fp16))[name = tensor("input_491_cast_fp16")]; tensor var_11164 = const()[name = tensor("op_11164"), val = tensor([1, 1])]; tensor var_11166 = const()[name = tensor("op_11166"), val = tensor([1, 1])]; tensor var_11168_pad_type_0 = const()[name = tensor("op_11168_pad_type_0"), val = tensor("custom")]; tensor var_11168_pad_0 = const()[name = tensor("op_11168_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(821638400)))]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(821843264)))]; tensor var_11168_cast_fp16 = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_11166, groups = var_10509, pad = var_11168_pad_0, pad_type = var_11168_pad_type_0, strides = var_11164, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_491_cast_fp16)[name = tensor("op_11168_cast_fp16")]; tensor inputs_87_cast_fp16 = add(x = var_11168_cast_fp16, y = inputs_85_cast_fp16)[name = tensor("inputs_87_cast_fp16")]; tensor var_11172 = const()[name = tensor("op_11172"), val = tensor([1])]; tensor channels_mean_87_cast_fp16 = reduce_mean(axes = var_11172, keep_dims = var_10504, x = inputs_87_cast_fp16)[name = tensor("channels_mean_87_cast_fp16")]; tensor zero_mean_87_cast_fp16 = sub(x = inputs_87_cast_fp16, y = channels_mean_87_cast_fp16)[name = tensor("zero_mean_87_cast_fp16")]; tensor zero_mean_sq_87_cast_fp16 = mul(x = zero_mean_87_cast_fp16, y = zero_mean_87_cast_fp16)[name = tensor("zero_mean_sq_87_cast_fp16")]; tensor var_11176 = const()[name = tensor("op_11176"), val = tensor([1])]; tensor var_11177_cast_fp16 = reduce_mean(axes = var_11176, keep_dims = var_10504, x = zero_mean_sq_87_cast_fp16)[name = tensor("op_11177_cast_fp16")]; tensor var_11178_to_fp16 = const()[name = tensor("op_11178_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_11179_cast_fp16 = add(x = var_11177_cast_fp16, y = var_11178_to_fp16)[name = tensor("op_11179_cast_fp16")]; tensor denom_87_epsilon_0_to_fp16 = const()[name = tensor("denom_87_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_87_cast_fp16 = rsqrt(epsilon = denom_87_epsilon_0_to_fp16, x = var_11179_cast_fp16)[name = tensor("denom_87_cast_fp16")]; tensor out_87_cast_fp16 = mul(x = zero_mean_87_cast_fp16, y = denom_87_cast_fp16)[name = tensor("out_87_cast_fp16")]; tensor var_11183_to_fp16 = const()[name = tensor("op_11183_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(821843968)))]; tensor var_11184_cast_fp16 = add(x = out_87_cast_fp16, y = var_11183_to_fp16)[name = tensor("op_11184_cast_fp16")]; tensor var_11186_to_fp16 = const()[name = tensor("op_11186_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(821844672)))]; tensor hidden_states_309_cast_fp16 = mul(x = var_11184_cast_fp16, y = var_11186_to_fp16)[name = tensor("hidden_states_309_cast_fp16")]; tensor var_11193 = const()[name = tensor("op_11193"), val = tensor([1, 1])]; tensor var_11195 = const()[name = tensor("op_11195"), val = tensor([1, 1])]; tensor q_59_pad_type_0 = const()[name = tensor("q_59_pad_type_0"), val = tensor("custom")]; tensor q_59_pad_0 = const()[name = tensor("q_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(821845376)))]; tensor q_59_cast_fp16 = conv(dilations = var_11195, groups = var_10509, pad = q_59_pad_0, pad_type = q_59_pad_type_0, strides = var_11193, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_309_cast_fp16)[name = tensor("q_59_cast_fp16")]; tensor var_11199 = const()[name = tensor("op_11199"), val = tensor([1, 1])]; tensor var_11201 = const()[name = tensor("op_11201"), val = tensor([1, 1])]; tensor k_117_pad_type_0 = const()[name = tensor("k_117_pad_type_0"), val = tensor("custom")]; tensor k_117_pad_0 = const()[name = tensor("k_117_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(822050240)))]; tensor k_117_cast_fp16 = conv(dilations = var_11201, groups = var_10509, pad = k_117_pad_0, pad_type = k_117_pad_type_0, strides = var_11199, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_117_cast_fp16")]; tensor var_11205 = const()[name = tensor("op_11205"), val = tensor([1, 1])]; tensor var_11207 = const()[name = tensor("op_11207"), val = tensor([1, 1])]; tensor v_59_pad_type_0 = const()[name = tensor("v_59_pad_type_0"), val = tensor("custom")]; tensor v_59_pad_0 = const()[name = tensor("v_59_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(822705664)))]; tensor v_59_cast_fp16 = conv(dilations = var_11207, groups = var_10509, pad = v_59_pad_0, pad_type = v_59_pad_type_0, strides = var_11205, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_59_cast_fp16")]; tensor var_11211_begin_0 = const()[name = tensor("op_11211_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11211_end_0 = const()[name = tensor("op_11211_end_0"), val = tensor([2, 64, 1, 2304])]; tensor var_11211_end_mask_0 = const()[name = tensor("op_11211_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11211_cast_fp16 = slice_by_index(begin = var_11211_begin_0, end = var_11211_end_0, end_mask = var_11211_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_11211_cast_fp16")]; tensor var_11215_begin_0 = const()[name = tensor("op_11215_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_11215_end_0 = const()[name = tensor("op_11215_end_0"), val = tensor([2, 128, 1, 2304])]; tensor var_11215_end_mask_0 = const()[name = tensor("op_11215_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11215_cast_fp16 = slice_by_index(begin = var_11215_begin_0, end = var_11215_end_0, end_mask = var_11215_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_11215_cast_fp16")]; tensor var_11219_begin_0 = const()[name = tensor("op_11219_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_11219_end_0 = const()[name = tensor("op_11219_end_0"), val = tensor([2, 192, 1, 2304])]; tensor var_11219_end_mask_0 = const()[name = tensor("op_11219_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11219_cast_fp16 = slice_by_index(begin = var_11219_begin_0, end = var_11219_end_0, end_mask = var_11219_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_11219_cast_fp16")]; tensor var_11223_begin_0 = const()[name = tensor("op_11223_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_11223_end_0 = const()[name = tensor("op_11223_end_0"), val = tensor([2, 256, 1, 2304])]; tensor var_11223_end_mask_0 = const()[name = tensor("op_11223_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11223_cast_fp16 = slice_by_index(begin = var_11223_begin_0, end = var_11223_end_0, end_mask = var_11223_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_11223_cast_fp16")]; tensor var_11227_begin_0 = const()[name = tensor("op_11227_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_11227_end_0 = const()[name = tensor("op_11227_end_0"), val = tensor([2, 320, 1, 2304])]; tensor var_11227_end_mask_0 = const()[name = tensor("op_11227_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11227_cast_fp16 = slice_by_index(begin = var_11227_begin_0, end = var_11227_end_0, end_mask = var_11227_end_mask_0, x = q_59_cast_fp16)[name = tensor("op_11227_cast_fp16")]; tensor k_119_perm_0 = const()[name = tensor("k_119_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_11234_begin_0 = const()[name = tensor("op_11234_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11234_end_0 = const()[name = tensor("op_11234_end_0"), val = tensor([2, 77, 1, 64])]; tensor var_11234_end_mask_0 = const()[name = tensor("op_11234_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_2 = transpose(perm = k_119_perm_0, x = k_117_cast_fp16)[name = tensor("transpose_2")]; tensor var_11234_cast_fp16 = slice_by_index(begin = var_11234_begin_0, end = var_11234_end_0, end_mask = var_11234_end_mask_0, x = transpose_2)[name = tensor("op_11234_cast_fp16")]; tensor var_11238_begin_0 = const()[name = tensor("op_11238_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_11238_end_0 = const()[name = tensor("op_11238_end_0"), val = tensor([2, 77, 1, 128])]; tensor var_11238_end_mask_0 = const()[name = tensor("op_11238_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11238_cast_fp16 = slice_by_index(begin = var_11238_begin_0, end = var_11238_end_0, end_mask = var_11238_end_mask_0, x = transpose_2)[name = tensor("op_11238_cast_fp16")]; tensor var_11242_begin_0 = const()[name = tensor("op_11242_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_11242_end_0 = const()[name = tensor("op_11242_end_0"), val = tensor([2, 77, 1, 192])]; tensor var_11242_end_mask_0 = const()[name = tensor("op_11242_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11242_cast_fp16 = slice_by_index(begin = var_11242_begin_0, end = var_11242_end_0, end_mask = var_11242_end_mask_0, x = transpose_2)[name = tensor("op_11242_cast_fp16")]; tensor var_11246_begin_0 = const()[name = tensor("op_11246_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_11246_end_0 = const()[name = tensor("op_11246_end_0"), val = tensor([2, 77, 1, 256])]; tensor var_11246_end_mask_0 = const()[name = tensor("op_11246_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11246_cast_fp16 = slice_by_index(begin = var_11246_begin_0, end = var_11246_end_0, end_mask = var_11246_end_mask_0, x = transpose_2)[name = tensor("op_11246_cast_fp16")]; tensor var_11250_begin_0 = const()[name = tensor("op_11250_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_11250_end_0 = const()[name = tensor("op_11250_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_11250_end_mask_0 = const()[name = tensor("op_11250_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11250_cast_fp16 = slice_by_index(begin = var_11250_begin_0, end = var_11250_end_0, end_mask = var_11250_end_mask_0, x = transpose_2)[name = tensor("op_11250_cast_fp16")]; tensor var_11252_begin_0 = const()[name = tensor("op_11252_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11252_end_0 = const()[name = tensor("op_11252_end_0"), val = tensor([2, 64, 1, 77])]; tensor var_11252_end_mask_0 = const()[name = tensor("op_11252_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11252_cast_fp16 = slice_by_index(begin = var_11252_begin_0, end = var_11252_end_0, end_mask = var_11252_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_11252_cast_fp16")]; tensor var_11256_begin_0 = const()[name = tensor("op_11256_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_11256_end_0 = const()[name = tensor("op_11256_end_0"), val = tensor([2, 128, 1, 77])]; tensor var_11256_end_mask_0 = const()[name = tensor("op_11256_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11256_cast_fp16 = slice_by_index(begin = var_11256_begin_0, end = var_11256_end_0, end_mask = var_11256_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_11256_cast_fp16")]; tensor var_11260_begin_0 = const()[name = tensor("op_11260_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_11260_end_0 = const()[name = tensor("op_11260_end_0"), val = tensor([2, 192, 1, 77])]; tensor var_11260_end_mask_0 = const()[name = tensor("op_11260_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11260_cast_fp16 = slice_by_index(begin = var_11260_begin_0, end = var_11260_end_0, end_mask = var_11260_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_11260_cast_fp16")]; tensor var_11264_begin_0 = const()[name = tensor("op_11264_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_11264_end_0 = const()[name = tensor("op_11264_end_0"), val = tensor([2, 256, 1, 77])]; tensor var_11264_end_mask_0 = const()[name = tensor("op_11264_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11264_cast_fp16 = slice_by_index(begin = var_11264_begin_0, end = var_11264_end_0, end_mask = var_11264_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_11264_cast_fp16")]; tensor var_11268_begin_0 = const()[name = tensor("op_11268_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_11268_end_0 = const()[name = tensor("op_11268_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_11268_end_mask_0 = const()[name = tensor("op_11268_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11268_cast_fp16 = slice_by_index(begin = var_11268_begin_0, end = var_11268_end_0, end_mask = var_11268_end_mask_0, x = v_59_cast_fp16)[name = tensor("op_11268_cast_fp16")]; tensor var_11272_equation_0 = const()[name = tensor("op_11272_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11272_cast_fp16 = einsum(equation = var_11272_equation_0, values = (var_11234_cast_fp16, var_11211_cast_fp16))[name = tensor("op_11272_cast_fp16")]; tensor var_11273_to_fp16 = const()[name = tensor("op_11273_to_fp16"), val = tensor(0x1p-3)]; tensor aw_751_cast_fp16 = mul(x = var_11272_cast_fp16, y = var_11273_to_fp16)[name = tensor("aw_751_cast_fp16")]; tensor var_11276_equation_0 = const()[name = tensor("op_11276_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11276_cast_fp16 = einsum(equation = var_11276_equation_0, values = (var_11238_cast_fp16, var_11215_cast_fp16))[name = tensor("op_11276_cast_fp16")]; tensor var_11277_to_fp16 = const()[name = tensor("op_11277_to_fp16"), val = tensor(0x1p-3)]; tensor aw_753_cast_fp16 = mul(x = var_11276_cast_fp16, y = var_11277_to_fp16)[name = tensor("aw_753_cast_fp16")]; tensor var_11280_equation_0 = const()[name = tensor("op_11280_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11280_cast_fp16 = einsum(equation = var_11280_equation_0, values = (var_11242_cast_fp16, var_11219_cast_fp16))[name = tensor("op_11280_cast_fp16")]; tensor var_11281_to_fp16 = const()[name = tensor("op_11281_to_fp16"), val = tensor(0x1p-3)]; tensor aw_755_cast_fp16 = mul(x = var_11280_cast_fp16, y = var_11281_to_fp16)[name = tensor("aw_755_cast_fp16")]; tensor var_11284_equation_0 = const()[name = tensor("op_11284_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11284_cast_fp16 = einsum(equation = var_11284_equation_0, values = (var_11246_cast_fp16, var_11223_cast_fp16))[name = tensor("op_11284_cast_fp16")]; tensor var_11285_to_fp16 = const()[name = tensor("op_11285_to_fp16"), val = tensor(0x1p-3)]; tensor aw_757_cast_fp16 = mul(x = var_11284_cast_fp16, y = var_11285_to_fp16)[name = tensor("aw_757_cast_fp16")]; tensor var_11288_equation_0 = const()[name = tensor("op_11288_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11288_cast_fp16 = einsum(equation = var_11288_equation_0, values = (var_11250_cast_fp16, var_11227_cast_fp16))[name = tensor("op_11288_cast_fp16")]; tensor var_11289_to_fp16 = const()[name = tensor("op_11289_to_fp16"), val = tensor(0x1p-3)]; tensor aw_759_cast_fp16 = mul(x = var_11288_cast_fp16, y = var_11289_to_fp16)[name = tensor("aw_759_cast_fp16")]; tensor var_11291_cast_fp16 = softmax(axis = var_10509, x = aw_751_cast_fp16)[name = tensor("op_11291_cast_fp16")]; tensor var_11292_cast_fp16 = softmax(axis = var_10509, x = aw_753_cast_fp16)[name = tensor("op_11292_cast_fp16")]; tensor var_11293_cast_fp16 = softmax(axis = var_10509, x = aw_755_cast_fp16)[name = tensor("op_11293_cast_fp16")]; tensor var_11294_cast_fp16 = softmax(axis = var_10509, x = aw_757_cast_fp16)[name = tensor("op_11294_cast_fp16")]; tensor var_11295_cast_fp16 = softmax(axis = var_10509, x = aw_759_cast_fp16)[name = tensor("op_11295_cast_fp16")]; tensor var_11297_equation_0 = const()[name = tensor("op_11297_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11297_cast_fp16 = einsum(equation = var_11297_equation_0, values = (var_11252_cast_fp16, var_11291_cast_fp16))[name = tensor("op_11297_cast_fp16")]; tensor var_11299_equation_0 = const()[name = tensor("op_11299_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11299_cast_fp16 = einsum(equation = var_11299_equation_0, values = (var_11256_cast_fp16, var_11292_cast_fp16))[name = tensor("op_11299_cast_fp16")]; tensor var_11301_equation_0 = const()[name = tensor("op_11301_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11301_cast_fp16 = einsum(equation = var_11301_equation_0, values = (var_11260_cast_fp16, var_11293_cast_fp16))[name = tensor("op_11301_cast_fp16")]; tensor var_11303_equation_0 = const()[name = tensor("op_11303_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11303_cast_fp16 = einsum(equation = var_11303_equation_0, values = (var_11264_cast_fp16, var_11294_cast_fp16))[name = tensor("op_11303_cast_fp16")]; tensor var_11305_equation_0 = const()[name = tensor("op_11305_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11305_cast_fp16 = einsum(equation = var_11305_equation_0, values = (var_11268_cast_fp16, var_11295_cast_fp16))[name = tensor("op_11305_cast_fp16")]; tensor input_493_interleave_0 = const()[name = tensor("input_493_interleave_0"), val = tensor(false)]; tensor input_493_cast_fp16 = concat(axis = var_10509, interleave = input_493_interleave_0, values = (var_11297_cast_fp16, var_11299_cast_fp16, var_11301_cast_fp16, var_11303_cast_fp16, var_11305_cast_fp16))[name = tensor("input_493_cast_fp16")]; tensor var_11311 = const()[name = tensor("op_11311"), val = tensor([1, 1])]; tensor var_11313 = const()[name = tensor("op_11313"), val = tensor([1, 1])]; tensor var_11315_pad_type_0 = const()[name = tensor("op_11315_pad_type_0"), val = tensor("custom")]; tensor var_11315_pad_0 = const()[name = tensor("op_11315_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(823361088)))]; tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(823565952)))]; tensor var_11315_cast_fp16 = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_11313, groups = var_10509, pad = var_11315_pad_0, pad_type = var_11315_pad_type_0, strides = var_11311, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_493_cast_fp16)[name = tensor("op_11315_cast_fp16")]; tensor inputs_89_cast_fp16 = add(x = var_11315_cast_fp16, y = inputs_87_cast_fp16)[name = tensor("inputs_89_cast_fp16")]; tensor var_11319 = const()[name = tensor("op_11319"), val = tensor([1])]; tensor channels_mean_89_cast_fp16 = reduce_mean(axes = var_11319, keep_dims = var_10504, x = inputs_89_cast_fp16)[name = tensor("channels_mean_89_cast_fp16")]; tensor zero_mean_89_cast_fp16 = sub(x = inputs_89_cast_fp16, y = channels_mean_89_cast_fp16)[name = tensor("zero_mean_89_cast_fp16")]; tensor zero_mean_sq_89_cast_fp16 = mul(x = zero_mean_89_cast_fp16, y = zero_mean_89_cast_fp16)[name = tensor("zero_mean_sq_89_cast_fp16")]; tensor var_11323 = const()[name = tensor("op_11323"), val = tensor([1])]; tensor var_11324_cast_fp16 = reduce_mean(axes = var_11323, keep_dims = var_10504, x = zero_mean_sq_89_cast_fp16)[name = tensor("op_11324_cast_fp16")]; tensor var_11325_to_fp16 = const()[name = tensor("op_11325_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_11326_cast_fp16 = add(x = var_11324_cast_fp16, y = var_11325_to_fp16)[name = tensor("op_11326_cast_fp16")]; tensor denom_89_epsilon_0_to_fp16 = const()[name = tensor("denom_89_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_89_cast_fp16 = rsqrt(epsilon = denom_89_epsilon_0_to_fp16, x = var_11326_cast_fp16)[name = tensor("denom_89_cast_fp16")]; tensor out_89_cast_fp16 = mul(x = zero_mean_89_cast_fp16, y = denom_89_cast_fp16)[name = tensor("out_89_cast_fp16")]; tensor var_11330_to_fp16 = const()[name = tensor("op_11330_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(823566656)))]; tensor var_11331_cast_fp16 = add(x = out_89_cast_fp16, y = var_11330_to_fp16)[name = tensor("op_11331_cast_fp16")]; tensor var_11333_to_fp16 = const()[name = tensor("op_11333_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(823567360)))]; tensor input_495_cast_fp16 = mul(x = var_11331_cast_fp16, y = var_11333_to_fp16)[name = tensor("input_495_cast_fp16")]; tensor var_11341 = const()[name = tensor("op_11341"), val = tensor([1, 1])]; tensor var_11343 = const()[name = tensor("op_11343"), val = tensor([1, 1])]; tensor var_11345_pad_type_0 = const()[name = tensor("op_11345_pad_type_0"), val = tensor("custom")]; tensor var_11345_pad_0 = const()[name = tensor("op_11345_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(823568064)))]; tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(825206528)))]; tensor var_11345_cast_fp16 = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_11343, groups = var_10509, pad = var_11345_pad_0, pad_type = var_11345_pad_type_0, strides = var_11341, weight = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_495_cast_fp16)[name = tensor("op_11345_cast_fp16")]; tensor var_11346_split_sizes_0 = const()[name = tensor("op_11346_split_sizes_0"), val = tensor([1280, 1280])]; tensor var_11346_axis_0 = const()[name = tensor("op_11346_axis_0"), val = tensor(1)]; tensor var_11346_cast_fp16_0, tensor var_11346_cast_fp16_1 = split(axis = var_11346_axis_0, split_sizes = var_11346_split_sizes_0, x = var_11345_cast_fp16)[name = tensor("op_11346_cast_fp16")]; tensor var_11348_mode_0 = const()[name = tensor("op_11348_mode_0"), val = tensor("EXACT")]; tensor var_11348_cast_fp16 = gelu(mode = var_11348_mode_0, x = var_11346_cast_fp16_1)[name = tensor("op_11348_cast_fp16")]; tensor input_497_cast_fp16 = mul(x = var_11346_cast_fp16_0, y = var_11348_cast_fp16)[name = tensor("input_497_cast_fp16")]; tensor var_11352 = const()[name = tensor("op_11352"), val = tensor([1, 1])]; tensor var_11354 = const()[name = tensor("op_11354"), val = tensor([1, 1])]; tensor var_11356_pad_type_0 = const()[name = tensor("op_11356_pad_type_0"), val = tensor("custom")]; tensor var_11356_pad_0 = const()[name = tensor("op_11356_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(825211712)))]; tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826030976)))]; tensor var_11356_cast_fp16 = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_11354, groups = var_10509, pad = var_11356_pad_0, pad_type = var_11356_pad_type_0, strides = var_11352, weight = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_497_cast_fp16)[name = tensor("op_11356_cast_fp16")]; tensor hidden_states_313_cast_fp16 = add(x = var_11356_cast_fp16, y = inputs_89_cast_fp16)[name = tensor("hidden_states_313_cast_fp16")]; tensor var_11358 = const()[name = tensor("op_11358"), val = tensor([2, 320, 48, 48])]; tensor input_499_cast_fp16 = reshape(shape = var_11358, x = hidden_states_313_cast_fp16)[name = tensor("input_499_cast_fp16")]; tensor var_11362 = const()[name = tensor("op_11362"), val = tensor([1, 1])]; tensor var_11364 = const()[name = tensor("op_11364"), val = tensor([1, 1])]; tensor hidden_states_315_pad_type_0 = const()[name = tensor("hidden_states_315_pad_type_0"), val = tensor("custom")]; tensor hidden_states_315_pad_0 = const()[name = tensor("hidden_states_315_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_1_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826031680)))]; tensor up_blocks_3_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826236544)))]; tensor hidden_states_315_cast_fp16 = conv(bias = up_blocks_3_attentions_1_proj_out_bias_to_fp16, dilations = var_11364, groups = var_10509, pad = hidden_states_315_pad_0, pad_type = hidden_states_315_pad_type_0, strides = var_11362, weight = up_blocks_3_attentions_1_proj_out_weight_to_fp16, x = input_499_cast_fp16)[name = tensor("hidden_states_315_cast_fp16")]; tensor hidden_states_317_cast_fp16 = add(x = hidden_states_315_cast_fp16, y = hidden_states_303_cast_fp16)[name = tensor("hidden_states_317_cast_fp16")]; tensor input_501_interleave_0 = const()[name = tensor("input_501_interleave_0"), val = tensor(false)]; tensor cast_13 = cast(dtype = cast_4_dtype_0, x = input_7_cast_fp16)[name = tensor("cast_13")]; tensor input_501_cast_fp16 = concat(axis = var_10509, interleave = input_501_interleave_0, values = (hidden_states_317_cast_fp16, cast_13))[name = tensor("input_501_cast_fp16")]; tensor reshape_228_shape_0 = const()[name = tensor("reshape_228_shape_0"), val = tensor([2, 32, 20, 48, 48])]; tensor reshape_228_cast_fp16 = reshape(shape = reshape_228_shape_0, x = input_501_cast_fp16)[name = tensor("reshape_228_cast_fp16")]; tensor reduce_mean_171_axes_0 = const()[name = tensor("reduce_mean_171_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_171_keep_dims_0 = const()[name = tensor("reduce_mean_171_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_171_cast_fp16 = reduce_mean(axes = reduce_mean_171_axes_0, keep_dims = reduce_mean_171_keep_dims_0, x = reshape_228_cast_fp16)[name = tensor("reduce_mean_171_cast_fp16")]; tensor sub_114_cast_fp16 = sub(x = reshape_228_cast_fp16, y = reduce_mean_171_cast_fp16)[name = tensor("sub_114_cast_fp16")]; tensor square_57_cast_fp16 = square(x = sub_114_cast_fp16)[name = tensor("square_57_cast_fp16")]; tensor reduce_mean_173_axes_0 = const()[name = tensor("reduce_mean_173_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_173_keep_dims_0 = const()[name = tensor("reduce_mean_173_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_173_cast_fp16 = reduce_mean(axes = reduce_mean_173_axes_0, keep_dims = reduce_mean_173_keep_dims_0, x = square_57_cast_fp16)[name = tensor("reduce_mean_173_cast_fp16")]; tensor add_114_y_0_to_fp16 = const()[name = tensor("add_114_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_114_cast_fp16 = add(x = reduce_mean_173_cast_fp16, y = add_114_y_0_to_fp16)[name = tensor("add_114_cast_fp16")]; tensor sqrt_57_cast_fp16 = sqrt(x = add_114_cast_fp16)[name = tensor("sqrt_57_cast_fp16")]; tensor real_div_57_cast_fp16 = real_div(x = sub_114_cast_fp16, y = sqrt_57_cast_fp16)[name = tensor("real_div_57_cast_fp16")]; tensor reshape_229_shape_0 = const()[name = tensor("reshape_229_shape_0"), val = tensor([2, 640, 48, 48])]; tensor reshape_229_cast_fp16 = reshape(shape = reshape_229_shape_0, x = real_div_57_cast_fp16)[name = tensor("reshape_229_cast_fp16")]; tensor add_115_gamma_0_to_fp16 = const()[name = tensor("add_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826237248)))]; tensor add_115_beta_0_to_fp16 = const()[name = tensor("add_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826238592)))]; tensor add_115_epsilon_0_to_fp16 = const()[name = tensor("add_115_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_115_cast_fp16 = batch_norm(beta = add_115_beta_0_to_fp16, epsilon = add_115_epsilon_0_to_fp16, gamma = add_115_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_229_cast_fp16)[name = tensor("add_115_cast_fp16")]; tensor input_505_cast_fp16 = silu(x = add_115_cast_fp16)[name = tensor("input_505_cast_fp16")]; tensor var_11382 = const()[name = tensor("op_11382"), val = tensor([1, 1])]; tensor var_11384 = const()[name = tensor("op_11384"), val = tensor([1, 1])]; tensor hidden_states_319_pad_type_0 = const()[name = tensor("hidden_states_319_pad_type_0"), val = tensor("custom")]; tensor hidden_states_319_pad_0 = const()[name = tensor("hidden_states_319_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_3_resnets_2_conv1_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(826239936)))]; tensor up_blocks_3_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(829926400)))]; tensor hidden_states_319_cast_fp16 = conv(bias = up_blocks_3_resnets_2_conv1_bias_to_fp16, dilations = var_11384, groups = var_10509, pad = hidden_states_319_pad_0, pad_type = hidden_states_319_pad_type_0, strides = var_11382, weight = up_blocks_3_resnets_2_conv1_weight_to_fp16, x = input_505_cast_fp16)[name = tensor("hidden_states_319_cast_fp16")]; tensor var_11390 = const()[name = tensor("op_11390"), val = tensor([1, 1])]; tensor var_11392 = const()[name = tensor("op_11392"), val = tensor([1, 1])]; tensor temb_pad_type_0 = const()[name = tensor("temb_pad_type_0"), val = tensor("custom")]; tensor temb_pad_0 = const()[name = tensor("temb_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(829927104)))]; tensor up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830746368)))]; tensor temb_cast_fp16 = conv(bias = up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_11392, groups = var_10509, pad = temb_pad_0, pad_type = temb_pad_type_0, strides = var_11390, weight = up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16, x = cast_12)[name = tensor("temb_cast_fp16")]; tensor input_509_cast_fp16 = add(x = hidden_states_319_cast_fp16, y = temb_cast_fp16)[name = tensor("input_509_cast_fp16")]; tensor reshape_232_shape_0 = const()[name = tensor("reshape_232_shape_0"), val = tensor([2, 32, 10, 48, 48])]; tensor reshape_232_cast_fp16 = reshape(shape = reshape_232_shape_0, x = input_509_cast_fp16)[name = tensor("reshape_232_cast_fp16")]; tensor reduce_mean_174_axes_0 = const()[name = tensor("reduce_mean_174_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_174_keep_dims_0 = const()[name = tensor("reduce_mean_174_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_174_cast_fp16 = reduce_mean(axes = reduce_mean_174_axes_0, keep_dims = reduce_mean_174_keep_dims_0, x = reshape_232_cast_fp16)[name = tensor("reduce_mean_174_cast_fp16")]; tensor sub_116_cast_fp16 = sub(x = reshape_232_cast_fp16, y = reduce_mean_174_cast_fp16)[name = tensor("sub_116_cast_fp16")]; tensor square_58_cast_fp16 = square(x = sub_116_cast_fp16)[name = tensor("square_58_cast_fp16")]; tensor reduce_mean_176_axes_0 = const()[name = tensor("reduce_mean_176_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_176_keep_dims_0 = const()[name = tensor("reduce_mean_176_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_176_cast_fp16 = reduce_mean(axes = reduce_mean_176_axes_0, keep_dims = reduce_mean_176_keep_dims_0, x = square_58_cast_fp16)[name = tensor("reduce_mean_176_cast_fp16")]; tensor add_116_y_0_to_fp16 = const()[name = tensor("add_116_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_116_cast_fp16 = add(x = reduce_mean_176_cast_fp16, y = add_116_y_0_to_fp16)[name = tensor("add_116_cast_fp16")]; tensor sqrt_58_cast_fp16 = sqrt(x = add_116_cast_fp16)[name = tensor("sqrt_58_cast_fp16")]; tensor real_div_58_cast_fp16 = real_div(x = sub_116_cast_fp16, y = sqrt_58_cast_fp16)[name = tensor("real_div_58_cast_fp16")]; tensor reshape_233_shape_0 = const()[name = tensor("reshape_233_shape_0"), val = tensor([2, 320, 48, 48])]; tensor reshape_233_cast_fp16 = reshape(shape = reshape_233_shape_0, x = real_div_58_cast_fp16)[name = tensor("reshape_233_cast_fp16")]; tensor add_117_gamma_0_to_fp16 = const()[name = tensor("add_117_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830747072)))]; tensor add_117_beta_0_to_fp16 = const()[name = tensor("add_117_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830747776)))]; tensor add_117_epsilon_0_to_fp16 = const()[name = tensor("add_117_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_117_cast_fp16 = batch_norm(beta = add_117_beta_0_to_fp16, epsilon = add_117_epsilon_0_to_fp16, gamma = add_117_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_233_cast_fp16)[name = tensor("add_117_cast_fp16")]; tensor input_513_cast_fp16 = silu(x = add_117_cast_fp16)[name = tensor("input_513_cast_fp16")]; tensor var_11402 = const()[name = tensor("op_11402"), val = tensor([1, 1])]; tensor var_11404 = const()[name = tensor("op_11404"), val = tensor([1, 1])]; tensor hidden_states_321_pad_type_0 = const()[name = tensor("hidden_states_321_pad_type_0"), val = tensor("custom")]; tensor hidden_states_321_pad_0 = const()[name = tensor("hidden_states_321_pad_0"), val = tensor([1, 1, 1, 1])]; tensor up_blocks_3_resnets_2_conv2_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(830748480)))]; tensor up_blocks_3_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(832591744)))]; tensor hidden_states_321_cast_fp16 = conv(bias = up_blocks_3_resnets_2_conv2_bias_to_fp16, dilations = var_11404, groups = var_10509, pad = hidden_states_321_pad_0, pad_type = hidden_states_321_pad_type_0, strides = var_11402, weight = up_blocks_3_resnets_2_conv2_weight_to_fp16, x = input_513_cast_fp16)[name = tensor("hidden_states_321_cast_fp16")]; tensor var_11409 = const()[name = tensor("op_11409"), val = tensor([1, 1])]; tensor var_11411 = const()[name = tensor("op_11411"), val = tensor([1, 1])]; tensor x_pad_type_0 = const()[name = tensor("x_pad_type_0"), val = tensor("custom")]; tensor x_pad_0 = const()[name = tensor("x_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(832592448)))]; tensor up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833002112)))]; tensor x_cast_fp16 = conv(bias = up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_11411, groups = var_10509, pad = x_pad_0, pad_type = x_pad_type_0, strides = var_11409, weight = up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16, x = input_501_cast_fp16)[name = tensor("x_cast_fp16")]; tensor hidden_states_323_cast_fp16 = add(x = x_cast_fp16, y = hidden_states_321_cast_fp16)[name = tensor("hidden_states_323_cast_fp16")]; tensor reshape_236_shape_0 = const()[name = tensor("reshape_236_shape_0"), val = tensor([2, 32, 10, 48, 48])]; tensor reshape_236_cast_fp16 = reshape(shape = reshape_236_shape_0, x = hidden_states_323_cast_fp16)[name = tensor("reshape_236_cast_fp16")]; tensor reduce_mean_177_axes_0 = const()[name = tensor("reduce_mean_177_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_177_keep_dims_0 = const()[name = tensor("reduce_mean_177_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_177_cast_fp16 = reduce_mean(axes = reduce_mean_177_axes_0, keep_dims = reduce_mean_177_keep_dims_0, x = reshape_236_cast_fp16)[name = tensor("reduce_mean_177_cast_fp16")]; tensor sub_118_cast_fp16 = sub(x = reshape_236_cast_fp16, y = reduce_mean_177_cast_fp16)[name = tensor("sub_118_cast_fp16")]; tensor square_59_cast_fp16 = square(x = sub_118_cast_fp16)[name = tensor("square_59_cast_fp16")]; tensor reduce_mean_179_axes_0 = const()[name = tensor("reduce_mean_179_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_179_keep_dims_0 = const()[name = tensor("reduce_mean_179_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_179_cast_fp16 = reduce_mean(axes = reduce_mean_179_axes_0, keep_dims = reduce_mean_179_keep_dims_0, x = square_59_cast_fp16)[name = tensor("reduce_mean_179_cast_fp16")]; tensor add_118_y_0_to_fp16 = const()[name = tensor("add_118_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_118_cast_fp16 = add(x = reduce_mean_179_cast_fp16, y = add_118_y_0_to_fp16)[name = tensor("add_118_cast_fp16")]; tensor sqrt_59_cast_fp16 = sqrt(x = add_118_cast_fp16)[name = tensor("sqrt_59_cast_fp16")]; tensor real_div_59_cast_fp16 = real_div(x = sub_118_cast_fp16, y = sqrt_59_cast_fp16)[name = tensor("real_div_59_cast_fp16")]; tensor reshape_237_shape_0 = const()[name = tensor("reshape_237_shape_0"), val = tensor([2, 320, 48, 48])]; tensor reshape_237_cast_fp16 = reshape(shape = reshape_237_shape_0, x = real_div_59_cast_fp16)[name = tensor("reshape_237_cast_fp16")]; tensor add_119_gamma_0_to_fp16 = const()[name = tensor("add_119_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833002816)))]; tensor add_119_beta_0_to_fp16 = const()[name = tensor("add_119_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833003520)))]; tensor add_119_epsilon_0_to_fp16 = const()[name = tensor("add_119_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_119_cast_fp16 = batch_norm(beta = add_119_beta_0_to_fp16, epsilon = add_119_epsilon_0_to_fp16, gamma = add_119_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_237_cast_fp16)[name = tensor("add_119_cast_fp16")]; tensor var_11431 = const()[name = tensor("op_11431"), val = tensor([1, 1])]; tensor var_11433 = const()[name = tensor("op_11433"), val = tensor([1, 1])]; tensor hidden_states_325_pad_type_0 = const()[name = tensor("hidden_states_325_pad_type_0"), val = tensor("custom")]; tensor hidden_states_325_pad_0 = const()[name = tensor("hidden_states_325_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_proj_in_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_proj_in_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833004224)))]; tensor up_blocks_3_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833209088)))]; tensor hidden_states_325_cast_fp16 = conv(bias = up_blocks_3_attentions_2_proj_in_bias_to_fp16, dilations = var_11433, groups = var_10509, pad = hidden_states_325_pad_0, pad_type = hidden_states_325_pad_type_0, strides = var_11431, weight = up_blocks_3_attentions_2_proj_in_weight_to_fp16, x = add_119_cast_fp16)[name = tensor("hidden_states_325_cast_fp16")]; tensor var_11438 = const()[name = tensor("op_11438"), val = tensor([2, 320, 1, 2304])]; tensor inputs_91_cast_fp16 = reshape(shape = var_11438, x = hidden_states_325_cast_fp16)[name = tensor("inputs_91_cast_fp16")]; tensor var_11448 = const()[name = tensor("op_11448"), val = tensor([1])]; tensor channels_mean_91_cast_fp16 = reduce_mean(axes = var_11448, keep_dims = var_10504, x = inputs_91_cast_fp16)[name = tensor("channels_mean_91_cast_fp16")]; tensor zero_mean_91_cast_fp16 = sub(x = inputs_91_cast_fp16, y = channels_mean_91_cast_fp16)[name = tensor("zero_mean_91_cast_fp16")]; tensor zero_mean_sq_91_cast_fp16 = mul(x = zero_mean_91_cast_fp16, y = zero_mean_91_cast_fp16)[name = tensor("zero_mean_sq_91_cast_fp16")]; tensor var_11452 = const()[name = tensor("op_11452"), val = tensor([1])]; tensor var_11453_cast_fp16 = reduce_mean(axes = var_11452, keep_dims = var_10504, x = zero_mean_sq_91_cast_fp16)[name = tensor("op_11453_cast_fp16")]; tensor var_11454_to_fp16 = const()[name = tensor("op_11454_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_11455_cast_fp16 = add(x = var_11453_cast_fp16, y = var_11454_to_fp16)[name = tensor("op_11455_cast_fp16")]; tensor denom_91_epsilon_0_to_fp16 = const()[name = tensor("denom_91_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_91_cast_fp16 = rsqrt(epsilon = denom_91_epsilon_0_to_fp16, x = var_11455_cast_fp16)[name = tensor("denom_91_cast_fp16")]; tensor out_91_cast_fp16 = mul(x = zero_mean_91_cast_fp16, y = denom_91_cast_fp16)[name = tensor("out_91_cast_fp16")]; tensor var_11459_to_fp16 = const()[name = tensor("op_11459_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833209792)))]; tensor var_11460_cast_fp16 = add(x = out_91_cast_fp16, y = var_11459_to_fp16)[name = tensor("op_11460_cast_fp16")]; tensor var_11462_to_fp16 = const()[name = tensor("op_11462_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833210496)))]; tensor hidden_states_327_cast_fp16 = mul(x = var_11460_cast_fp16, y = var_11462_to_fp16)[name = tensor("hidden_states_327_cast_fp16")]; tensor var_11469 = const()[name = tensor("op_11469"), val = tensor([1, 1])]; tensor var_11471 = const()[name = tensor("op_11471"), val = tensor([1, 1])]; tensor q_61_pad_type_0 = const()[name = tensor("q_61_pad_type_0"), val = tensor("custom")]; tensor q_61_pad_0 = const()[name = tensor("q_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833211200)))]; tensor q_61_cast_fp16 = conv(dilations = var_11471, groups = var_10509, pad = q_61_pad_0, pad_type = q_61_pad_type_0, strides = var_11469, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16, x = hidden_states_327_cast_fp16)[name = tensor("q_61_cast_fp16")]; tensor var_11475 = const()[name = tensor("op_11475"), val = tensor([1, 1])]; tensor var_11477 = const()[name = tensor("op_11477"), val = tensor([1, 1])]; tensor k_121_pad_type_0 = const()[name = tensor("k_121_pad_type_0"), val = tensor("custom")]; tensor k_121_pad_0 = const()[name = tensor("k_121_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833416064)))]; tensor k_121_cast_fp16 = conv(dilations = var_11477, groups = var_10509, pad = k_121_pad_0, pad_type = k_121_pad_type_0, strides = var_11475, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16, x = hidden_states_327_cast_fp16)[name = tensor("k_121_cast_fp16")]; tensor var_11481 = const()[name = tensor("op_11481"), val = tensor([1, 1])]; tensor var_11483 = const()[name = tensor("op_11483"), val = tensor([1, 1])]; tensor v_61_pad_type_0 = const()[name = tensor("v_61_pad_type_0"), val = tensor("custom")]; tensor v_61_pad_0 = const()[name = tensor("v_61_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833620928)))]; tensor v_61_cast_fp16 = conv(dilations = var_11483, groups = var_10509, pad = v_61_pad_0, pad_type = v_61_pad_type_0, strides = var_11481, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16, x = hidden_states_327_cast_fp16)[name = tensor("v_61_cast_fp16")]; tensor var_11487_begin_0 = const()[name = tensor("op_11487_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11487_end_0 = const()[name = tensor("op_11487_end_0"), val = tensor([2, 64, 1, 2304])]; tensor var_11487_end_mask_0 = const()[name = tensor("op_11487_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11487_cast_fp16 = slice_by_index(begin = var_11487_begin_0, end = var_11487_end_0, end_mask = var_11487_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_11487_cast_fp16")]; tensor var_11491_begin_0 = const()[name = tensor("op_11491_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_11491_end_0 = const()[name = tensor("op_11491_end_0"), val = tensor([2, 128, 1, 2304])]; tensor var_11491_end_mask_0 = const()[name = tensor("op_11491_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11491_cast_fp16 = slice_by_index(begin = var_11491_begin_0, end = var_11491_end_0, end_mask = var_11491_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_11491_cast_fp16")]; tensor var_11495_begin_0 = const()[name = tensor("op_11495_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_11495_end_0 = const()[name = tensor("op_11495_end_0"), val = tensor([2, 192, 1, 2304])]; tensor var_11495_end_mask_0 = const()[name = tensor("op_11495_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11495_cast_fp16 = slice_by_index(begin = var_11495_begin_0, end = var_11495_end_0, end_mask = var_11495_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_11495_cast_fp16")]; tensor var_11499_begin_0 = const()[name = tensor("op_11499_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_11499_end_0 = const()[name = tensor("op_11499_end_0"), val = tensor([2, 256, 1, 2304])]; tensor var_11499_end_mask_0 = const()[name = tensor("op_11499_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11499_cast_fp16 = slice_by_index(begin = var_11499_begin_0, end = var_11499_end_0, end_mask = var_11499_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_11499_cast_fp16")]; tensor var_11503_begin_0 = const()[name = tensor("op_11503_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_11503_end_0 = const()[name = tensor("op_11503_end_0"), val = tensor([2, 320, 1, 2304])]; tensor var_11503_end_mask_0 = const()[name = tensor("op_11503_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11503_cast_fp16 = slice_by_index(begin = var_11503_begin_0, end = var_11503_end_0, end_mask = var_11503_end_mask_0, x = q_61_cast_fp16)[name = tensor("op_11503_cast_fp16")]; tensor k_123_perm_0 = const()[name = tensor("k_123_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_11510_begin_0 = const()[name = tensor("op_11510_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11510_end_0 = const()[name = tensor("op_11510_end_0"), val = tensor([2, 2304, 1, 64])]; tensor var_11510_end_mask_0 = const()[name = tensor("op_11510_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_1 = transpose(perm = k_123_perm_0, x = k_121_cast_fp16)[name = tensor("transpose_1")]; tensor var_11510_cast_fp16 = slice_by_index(begin = var_11510_begin_0, end = var_11510_end_0, end_mask = var_11510_end_mask_0, x = transpose_1)[name = tensor("op_11510_cast_fp16")]; tensor var_11514_begin_0 = const()[name = tensor("op_11514_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_11514_end_0 = const()[name = tensor("op_11514_end_0"), val = tensor([2, 2304, 1, 128])]; tensor var_11514_end_mask_0 = const()[name = tensor("op_11514_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11514_cast_fp16 = slice_by_index(begin = var_11514_begin_0, end = var_11514_end_0, end_mask = var_11514_end_mask_0, x = transpose_1)[name = tensor("op_11514_cast_fp16")]; tensor var_11518_begin_0 = const()[name = tensor("op_11518_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_11518_end_0 = const()[name = tensor("op_11518_end_0"), val = tensor([2, 2304, 1, 192])]; tensor var_11518_end_mask_0 = const()[name = tensor("op_11518_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11518_cast_fp16 = slice_by_index(begin = var_11518_begin_0, end = var_11518_end_0, end_mask = var_11518_end_mask_0, x = transpose_1)[name = tensor("op_11518_cast_fp16")]; tensor var_11522_begin_0 = const()[name = tensor("op_11522_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_11522_end_0 = const()[name = tensor("op_11522_end_0"), val = tensor([2, 2304, 1, 256])]; tensor var_11522_end_mask_0 = const()[name = tensor("op_11522_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11522_cast_fp16 = slice_by_index(begin = var_11522_begin_0, end = var_11522_end_0, end_mask = var_11522_end_mask_0, x = transpose_1)[name = tensor("op_11522_cast_fp16")]; tensor var_11526_begin_0 = const()[name = tensor("op_11526_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_11526_end_0 = const()[name = tensor("op_11526_end_0"), val = tensor([2, 2304, 1, 320])]; tensor var_11526_end_mask_0 = const()[name = tensor("op_11526_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11526_cast_fp16 = slice_by_index(begin = var_11526_begin_0, end = var_11526_end_0, end_mask = var_11526_end_mask_0, x = transpose_1)[name = tensor("op_11526_cast_fp16")]; tensor var_11528_begin_0 = const()[name = tensor("op_11528_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11528_end_0 = const()[name = tensor("op_11528_end_0"), val = tensor([2, 64, 1, 2304])]; tensor var_11528_end_mask_0 = const()[name = tensor("op_11528_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11528_cast_fp16 = slice_by_index(begin = var_11528_begin_0, end = var_11528_end_0, end_mask = var_11528_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_11528_cast_fp16")]; tensor var_11532_begin_0 = const()[name = tensor("op_11532_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_11532_end_0 = const()[name = tensor("op_11532_end_0"), val = tensor([2, 128, 1, 2304])]; tensor var_11532_end_mask_0 = const()[name = tensor("op_11532_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11532_cast_fp16 = slice_by_index(begin = var_11532_begin_0, end = var_11532_end_0, end_mask = var_11532_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_11532_cast_fp16")]; tensor var_11536_begin_0 = const()[name = tensor("op_11536_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_11536_end_0 = const()[name = tensor("op_11536_end_0"), val = tensor([2, 192, 1, 2304])]; tensor var_11536_end_mask_0 = const()[name = tensor("op_11536_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11536_cast_fp16 = slice_by_index(begin = var_11536_begin_0, end = var_11536_end_0, end_mask = var_11536_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_11536_cast_fp16")]; tensor var_11540_begin_0 = const()[name = tensor("op_11540_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_11540_end_0 = const()[name = tensor("op_11540_end_0"), val = tensor([2, 256, 1, 2304])]; tensor var_11540_end_mask_0 = const()[name = tensor("op_11540_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11540_cast_fp16 = slice_by_index(begin = var_11540_begin_0, end = var_11540_end_0, end_mask = var_11540_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_11540_cast_fp16")]; tensor var_11544_begin_0 = const()[name = tensor("op_11544_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_11544_end_0 = const()[name = tensor("op_11544_end_0"), val = tensor([2, 320, 1, 2304])]; tensor var_11544_end_mask_0 = const()[name = tensor("op_11544_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11544_cast_fp16 = slice_by_index(begin = var_11544_begin_0, end = var_11544_end_0, end_mask = var_11544_end_mask_0, x = v_61_cast_fp16)[name = tensor("op_11544_cast_fp16")]; tensor var_11548_equation_0 = const()[name = tensor("op_11548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11548_cast_fp16 = einsum(equation = var_11548_equation_0, values = (var_11510_cast_fp16, var_11487_cast_fp16))[name = tensor("op_11548_cast_fp16")]; tensor var_11549_to_fp16 = const()[name = tensor("op_11549_to_fp16"), val = tensor(0x1p-3)]; tensor aw_761_cast_fp16 = mul(x = var_11548_cast_fp16, y = var_11549_to_fp16)[name = tensor("aw_761_cast_fp16")]; tensor var_11552_equation_0 = const()[name = tensor("op_11552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11552_cast_fp16 = einsum(equation = var_11552_equation_0, values = (var_11514_cast_fp16, var_11491_cast_fp16))[name = tensor("op_11552_cast_fp16")]; tensor var_11553_to_fp16 = const()[name = tensor("op_11553_to_fp16"), val = tensor(0x1p-3)]; tensor aw_763_cast_fp16 = mul(x = var_11552_cast_fp16, y = var_11553_to_fp16)[name = tensor("aw_763_cast_fp16")]; tensor var_11556_equation_0 = const()[name = tensor("op_11556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11556_cast_fp16 = einsum(equation = var_11556_equation_0, values = (var_11518_cast_fp16, var_11495_cast_fp16))[name = tensor("op_11556_cast_fp16")]; tensor var_11557_to_fp16 = const()[name = tensor("op_11557_to_fp16"), val = tensor(0x1p-3)]; tensor aw_765_cast_fp16 = mul(x = var_11556_cast_fp16, y = var_11557_to_fp16)[name = tensor("aw_765_cast_fp16")]; tensor var_11560_equation_0 = const()[name = tensor("op_11560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11560_cast_fp16 = einsum(equation = var_11560_equation_0, values = (var_11522_cast_fp16, var_11499_cast_fp16))[name = tensor("op_11560_cast_fp16")]; tensor var_11561_to_fp16 = const()[name = tensor("op_11561_to_fp16"), val = tensor(0x1p-3)]; tensor aw_767_cast_fp16 = mul(x = var_11560_cast_fp16, y = var_11561_to_fp16)[name = tensor("aw_767_cast_fp16")]; tensor var_11564_equation_0 = const()[name = tensor("op_11564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11564_cast_fp16 = einsum(equation = var_11564_equation_0, values = (var_11526_cast_fp16, var_11503_cast_fp16))[name = tensor("op_11564_cast_fp16")]; tensor var_11565_to_fp16 = const()[name = tensor("op_11565_to_fp16"), val = tensor(0x1p-3)]; tensor aw_769_cast_fp16 = mul(x = var_11564_cast_fp16, y = var_11565_to_fp16)[name = tensor("aw_769_cast_fp16")]; tensor var_11567_cast_fp16 = softmax(axis = var_10509, x = aw_761_cast_fp16)[name = tensor("op_11567_cast_fp16")]; tensor var_11568_cast_fp16 = softmax(axis = var_10509, x = aw_763_cast_fp16)[name = tensor("op_11568_cast_fp16")]; tensor var_11569_cast_fp16 = softmax(axis = var_10509, x = aw_765_cast_fp16)[name = tensor("op_11569_cast_fp16")]; tensor var_11570_cast_fp16 = softmax(axis = var_10509, x = aw_767_cast_fp16)[name = tensor("op_11570_cast_fp16")]; tensor var_11571_cast_fp16 = softmax(axis = var_10509, x = aw_769_cast_fp16)[name = tensor("op_11571_cast_fp16")]; tensor var_11573_equation_0 = const()[name = tensor("op_11573_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11573_cast_fp16 = einsum(equation = var_11573_equation_0, values = (var_11528_cast_fp16, var_11567_cast_fp16))[name = tensor("op_11573_cast_fp16")]; tensor var_11575_equation_0 = const()[name = tensor("op_11575_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11575_cast_fp16 = einsum(equation = var_11575_equation_0, values = (var_11532_cast_fp16, var_11568_cast_fp16))[name = tensor("op_11575_cast_fp16")]; tensor var_11577_equation_0 = const()[name = tensor("op_11577_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11577_cast_fp16 = einsum(equation = var_11577_equation_0, values = (var_11536_cast_fp16, var_11569_cast_fp16))[name = tensor("op_11577_cast_fp16")]; tensor var_11579_equation_0 = const()[name = tensor("op_11579_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11579_cast_fp16 = einsum(equation = var_11579_equation_0, values = (var_11540_cast_fp16, var_11570_cast_fp16))[name = tensor("op_11579_cast_fp16")]; tensor var_11581_equation_0 = const()[name = tensor("op_11581_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11581_cast_fp16 = einsum(equation = var_11581_equation_0, values = (var_11544_cast_fp16, var_11571_cast_fp16))[name = tensor("op_11581_cast_fp16")]; tensor input_517_interleave_0 = const()[name = tensor("input_517_interleave_0"), val = tensor(false)]; tensor input_517_cast_fp16 = concat(axis = var_10509, interleave = input_517_interleave_0, values = (var_11573_cast_fp16, var_11575_cast_fp16, var_11577_cast_fp16, var_11579_cast_fp16, var_11581_cast_fp16))[name = tensor("input_517_cast_fp16")]; tensor var_11587 = const()[name = tensor("op_11587"), val = tensor([1, 1])]; tensor var_11589 = const()[name = tensor("op_11589"), val = tensor([1, 1])]; tensor var_11591_pad_type_0 = const()[name = tensor("op_11591_pad_type_0"), val = tensor("custom")]; tensor var_11591_pad_0 = const()[name = tensor("op_11591_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(833825792)))]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(834030656)))]; tensor var_11591_cast_fp16 = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_11589, groups = var_10509, pad = var_11591_pad_0, pad_type = var_11591_pad_type_0, strides = var_11587, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16, x = input_517_cast_fp16)[name = tensor("op_11591_cast_fp16")]; tensor inputs_93_cast_fp16 = add(x = var_11591_cast_fp16, y = inputs_91_cast_fp16)[name = tensor("inputs_93_cast_fp16")]; tensor var_11595 = const()[name = tensor("op_11595"), val = tensor([1])]; tensor channels_mean_93_cast_fp16 = reduce_mean(axes = var_11595, keep_dims = var_10504, x = inputs_93_cast_fp16)[name = tensor("channels_mean_93_cast_fp16")]; tensor zero_mean_93_cast_fp16 = sub(x = inputs_93_cast_fp16, y = channels_mean_93_cast_fp16)[name = tensor("zero_mean_93_cast_fp16")]; tensor zero_mean_sq_93_cast_fp16 = mul(x = zero_mean_93_cast_fp16, y = zero_mean_93_cast_fp16)[name = tensor("zero_mean_sq_93_cast_fp16")]; tensor var_11599 = const()[name = tensor("op_11599"), val = tensor([1])]; tensor var_11600_cast_fp16 = reduce_mean(axes = var_11599, keep_dims = var_10504, x = zero_mean_sq_93_cast_fp16)[name = tensor("op_11600_cast_fp16")]; tensor var_11601_to_fp16 = const()[name = tensor("op_11601_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_11602_cast_fp16 = add(x = var_11600_cast_fp16, y = var_11601_to_fp16)[name = tensor("op_11602_cast_fp16")]; tensor denom_93_epsilon_0_to_fp16 = const()[name = tensor("denom_93_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_93_cast_fp16 = rsqrt(epsilon = denom_93_epsilon_0_to_fp16, x = var_11602_cast_fp16)[name = tensor("denom_93_cast_fp16")]; tensor out_93_cast_fp16 = mul(x = zero_mean_93_cast_fp16, y = denom_93_cast_fp16)[name = tensor("out_93_cast_fp16")]; tensor var_11606_to_fp16 = const()[name = tensor("op_11606_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(834031360)))]; tensor var_11607_cast_fp16 = add(x = out_93_cast_fp16, y = var_11606_to_fp16)[name = tensor("op_11607_cast_fp16")]; tensor var_11609_to_fp16 = const()[name = tensor("op_11609_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(834032064)))]; tensor hidden_states_329_cast_fp16 = mul(x = var_11607_cast_fp16, y = var_11609_to_fp16)[name = tensor("hidden_states_329_cast_fp16")]; tensor var_11616 = const()[name = tensor("op_11616"), val = tensor([1, 1])]; tensor var_11618 = const()[name = tensor("op_11618"), val = tensor([1, 1])]; tensor q_pad_type_0 = const()[name = tensor("q_pad_type_0"), val = tensor("custom")]; tensor q_pad_0 = const()[name = tensor("q_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(834032768)))]; tensor q_cast_fp16 = conv(dilations = var_11618, groups = var_10509, pad = q_pad_0, pad_type = q_pad_type_0, strides = var_11616, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16, x = hidden_states_329_cast_fp16)[name = tensor("q_cast_fp16")]; tensor var_11622 = const()[name = tensor("op_11622"), val = tensor([1, 1])]; tensor var_11624 = const()[name = tensor("op_11624"), val = tensor([1, 1])]; tensor k_125_pad_type_0 = const()[name = tensor("k_125_pad_type_0"), val = tensor("custom")]; tensor k_125_pad_0 = const()[name = tensor("k_125_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(834237632)))]; tensor k_125_cast_fp16 = conv(dilations = var_11624, groups = var_10509, pad = k_125_pad_0, pad_type = k_125_pad_type_0, strides = var_11622, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16, x = encoder_hidden_states)[name = tensor("k_125_cast_fp16")]; tensor var_11628 = const()[name = tensor("op_11628"), val = tensor([1, 1])]; tensor var_11630 = const()[name = tensor("op_11630"), val = tensor([1, 1])]; tensor v_pad_type_0 = const()[name = tensor("v_pad_type_0"), val = tensor("custom")]; tensor v_pad_0 = const()[name = tensor("v_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(834893056)))]; tensor v_cast_fp16 = conv(dilations = var_11630, groups = var_10509, pad = v_pad_0, pad_type = v_pad_type_0, strides = var_11628, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16, x = encoder_hidden_states)[name = tensor("v_cast_fp16")]; tensor var_11634_begin_0 = const()[name = tensor("op_11634_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11634_end_0 = const()[name = tensor("op_11634_end_0"), val = tensor([2, 64, 1, 2304])]; tensor var_11634_end_mask_0 = const()[name = tensor("op_11634_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11634_cast_fp16 = slice_by_index(begin = var_11634_begin_0, end = var_11634_end_0, end_mask = var_11634_end_mask_0, x = q_cast_fp16)[name = tensor("op_11634_cast_fp16")]; tensor var_11638_begin_0 = const()[name = tensor("op_11638_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_11638_end_0 = const()[name = tensor("op_11638_end_0"), val = tensor([2, 128, 1, 2304])]; tensor var_11638_end_mask_0 = const()[name = tensor("op_11638_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11638_cast_fp16 = slice_by_index(begin = var_11638_begin_0, end = var_11638_end_0, end_mask = var_11638_end_mask_0, x = q_cast_fp16)[name = tensor("op_11638_cast_fp16")]; tensor var_11642_begin_0 = const()[name = tensor("op_11642_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_11642_end_0 = const()[name = tensor("op_11642_end_0"), val = tensor([2, 192, 1, 2304])]; tensor var_11642_end_mask_0 = const()[name = tensor("op_11642_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11642_cast_fp16 = slice_by_index(begin = var_11642_begin_0, end = var_11642_end_0, end_mask = var_11642_end_mask_0, x = q_cast_fp16)[name = tensor("op_11642_cast_fp16")]; tensor var_11646_begin_0 = const()[name = tensor("op_11646_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_11646_end_0 = const()[name = tensor("op_11646_end_0"), val = tensor([2, 256, 1, 2304])]; tensor var_11646_end_mask_0 = const()[name = tensor("op_11646_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11646_cast_fp16 = slice_by_index(begin = var_11646_begin_0, end = var_11646_end_0, end_mask = var_11646_end_mask_0, x = q_cast_fp16)[name = tensor("op_11646_cast_fp16")]; tensor var_11650_begin_0 = const()[name = tensor("op_11650_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_11650_end_0 = const()[name = tensor("op_11650_end_0"), val = tensor([2, 320, 1, 2304])]; tensor var_11650_end_mask_0 = const()[name = tensor("op_11650_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11650_cast_fp16 = slice_by_index(begin = var_11650_begin_0, end = var_11650_end_0, end_mask = var_11650_end_mask_0, x = q_cast_fp16)[name = tensor("op_11650_cast_fp16")]; tensor k_perm_0 = const()[name = tensor("k_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_11657_begin_0 = const()[name = tensor("op_11657_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11657_end_0 = const()[name = tensor("op_11657_end_0"), val = tensor([2, 77, 1, 64])]; tensor var_11657_end_mask_0 = const()[name = tensor("op_11657_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_0 = transpose(perm = k_perm_0, x = k_125_cast_fp16)[name = tensor("transpose_0")]; tensor var_11657_cast_fp16 = slice_by_index(begin = var_11657_begin_0, end = var_11657_end_0, end_mask = var_11657_end_mask_0, x = transpose_0)[name = tensor("op_11657_cast_fp16")]; tensor var_11661_begin_0 = const()[name = tensor("op_11661_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_11661_end_0 = const()[name = tensor("op_11661_end_0"), val = tensor([2, 77, 1, 128])]; tensor var_11661_end_mask_0 = const()[name = tensor("op_11661_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11661_cast_fp16 = slice_by_index(begin = var_11661_begin_0, end = var_11661_end_0, end_mask = var_11661_end_mask_0, x = transpose_0)[name = tensor("op_11661_cast_fp16")]; tensor var_11665_begin_0 = const()[name = tensor("op_11665_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_11665_end_0 = const()[name = tensor("op_11665_end_0"), val = tensor([2, 77, 1, 192])]; tensor var_11665_end_mask_0 = const()[name = tensor("op_11665_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11665_cast_fp16 = slice_by_index(begin = var_11665_begin_0, end = var_11665_end_0, end_mask = var_11665_end_mask_0, x = transpose_0)[name = tensor("op_11665_cast_fp16")]; tensor var_11669_begin_0 = const()[name = tensor("op_11669_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_11669_end_0 = const()[name = tensor("op_11669_end_0"), val = tensor([2, 77, 1, 256])]; tensor var_11669_end_mask_0 = const()[name = tensor("op_11669_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11669_cast_fp16 = slice_by_index(begin = var_11669_begin_0, end = var_11669_end_0, end_mask = var_11669_end_mask_0, x = transpose_0)[name = tensor("op_11669_cast_fp16")]; tensor var_11673_begin_0 = const()[name = tensor("op_11673_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_11673_end_0 = const()[name = tensor("op_11673_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_11673_end_mask_0 = const()[name = tensor("op_11673_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11673_cast_fp16 = slice_by_index(begin = var_11673_begin_0, end = var_11673_end_0, end_mask = var_11673_end_mask_0, x = transpose_0)[name = tensor("op_11673_cast_fp16")]; tensor var_11675_begin_0 = const()[name = tensor("op_11675_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11675_end_0 = const()[name = tensor("op_11675_end_0"), val = tensor([2, 64, 1, 77])]; tensor var_11675_end_mask_0 = const()[name = tensor("op_11675_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11675_cast_fp16 = slice_by_index(begin = var_11675_begin_0, end = var_11675_end_0, end_mask = var_11675_end_mask_0, x = v_cast_fp16)[name = tensor("op_11675_cast_fp16")]; tensor var_11679_begin_0 = const()[name = tensor("op_11679_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_11679_end_0 = const()[name = tensor("op_11679_end_0"), val = tensor([2, 128, 1, 77])]; tensor var_11679_end_mask_0 = const()[name = tensor("op_11679_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11679_cast_fp16 = slice_by_index(begin = var_11679_begin_0, end = var_11679_end_0, end_mask = var_11679_end_mask_0, x = v_cast_fp16)[name = tensor("op_11679_cast_fp16")]; tensor var_11683_begin_0 = const()[name = tensor("op_11683_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_11683_end_0 = const()[name = tensor("op_11683_end_0"), val = tensor([2, 192, 1, 77])]; tensor var_11683_end_mask_0 = const()[name = tensor("op_11683_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11683_cast_fp16 = slice_by_index(begin = var_11683_begin_0, end = var_11683_end_0, end_mask = var_11683_end_mask_0, x = v_cast_fp16)[name = tensor("op_11683_cast_fp16")]; tensor var_11687_begin_0 = const()[name = tensor("op_11687_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_11687_end_0 = const()[name = tensor("op_11687_end_0"), val = tensor([2, 256, 1, 77])]; tensor var_11687_end_mask_0 = const()[name = tensor("op_11687_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11687_cast_fp16 = slice_by_index(begin = var_11687_begin_0, end = var_11687_end_0, end_mask = var_11687_end_mask_0, x = v_cast_fp16)[name = tensor("op_11687_cast_fp16")]; tensor var_11691_begin_0 = const()[name = tensor("op_11691_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_11691_end_0 = const()[name = tensor("op_11691_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_11691_end_mask_0 = const()[name = tensor("op_11691_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11691_cast_fp16 = slice_by_index(begin = var_11691_begin_0, end = var_11691_end_0, end_mask = var_11691_end_mask_0, x = v_cast_fp16)[name = tensor("op_11691_cast_fp16")]; tensor var_11695_equation_0 = const()[name = tensor("op_11695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11695_cast_fp16 = einsum(equation = var_11695_equation_0, values = (var_11657_cast_fp16, var_11634_cast_fp16))[name = tensor("op_11695_cast_fp16")]; tensor var_11696_to_fp16 = const()[name = tensor("op_11696_to_fp16"), val = tensor(0x1p-3)]; tensor aw_771_cast_fp16 = mul(x = var_11695_cast_fp16, y = var_11696_to_fp16)[name = tensor("aw_771_cast_fp16")]; tensor var_11699_equation_0 = const()[name = tensor("op_11699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11699_cast_fp16 = einsum(equation = var_11699_equation_0, values = (var_11661_cast_fp16, var_11638_cast_fp16))[name = tensor("op_11699_cast_fp16")]; tensor var_11700_to_fp16 = const()[name = tensor("op_11700_to_fp16"), val = tensor(0x1p-3)]; tensor aw_773_cast_fp16 = mul(x = var_11699_cast_fp16, y = var_11700_to_fp16)[name = tensor("aw_773_cast_fp16")]; tensor var_11703_equation_0 = const()[name = tensor("op_11703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11703_cast_fp16 = einsum(equation = var_11703_equation_0, values = (var_11665_cast_fp16, var_11642_cast_fp16))[name = tensor("op_11703_cast_fp16")]; tensor var_11704_to_fp16 = const()[name = tensor("op_11704_to_fp16"), val = tensor(0x1p-3)]; tensor aw_775_cast_fp16 = mul(x = var_11703_cast_fp16, y = var_11704_to_fp16)[name = tensor("aw_775_cast_fp16")]; tensor var_11707_equation_0 = const()[name = tensor("op_11707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11707_cast_fp16 = einsum(equation = var_11707_equation_0, values = (var_11669_cast_fp16, var_11646_cast_fp16))[name = tensor("op_11707_cast_fp16")]; tensor var_11708_to_fp16 = const()[name = tensor("op_11708_to_fp16"), val = tensor(0x1p-3)]; tensor aw_777_cast_fp16 = mul(x = var_11707_cast_fp16, y = var_11708_to_fp16)[name = tensor("aw_777_cast_fp16")]; tensor var_11711_equation_0 = const()[name = tensor("op_11711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11711_cast_fp16 = einsum(equation = var_11711_equation_0, values = (var_11673_cast_fp16, var_11650_cast_fp16))[name = tensor("op_11711_cast_fp16")]; tensor var_11712_to_fp16 = const()[name = tensor("op_11712_to_fp16"), val = tensor(0x1p-3)]; tensor aw_cast_fp16 = mul(x = var_11711_cast_fp16, y = var_11712_to_fp16)[name = tensor("aw_cast_fp16")]; tensor var_11714_cast_fp16 = softmax(axis = var_10509, x = aw_771_cast_fp16)[name = tensor("op_11714_cast_fp16")]; tensor var_11715_cast_fp16 = softmax(axis = var_10509, x = aw_773_cast_fp16)[name = tensor("op_11715_cast_fp16")]; tensor var_11716_cast_fp16 = softmax(axis = var_10509, x = aw_775_cast_fp16)[name = tensor("op_11716_cast_fp16")]; tensor var_11717_cast_fp16 = softmax(axis = var_10509, x = aw_777_cast_fp16)[name = tensor("op_11717_cast_fp16")]; tensor var_11718_cast_fp16 = softmax(axis = var_10509, x = aw_cast_fp16)[name = tensor("op_11718_cast_fp16")]; tensor var_11720_equation_0 = const()[name = tensor("op_11720_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11720_cast_fp16 = einsum(equation = var_11720_equation_0, values = (var_11675_cast_fp16, var_11714_cast_fp16))[name = tensor("op_11720_cast_fp16")]; tensor var_11722_equation_0 = const()[name = tensor("op_11722_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11722_cast_fp16 = einsum(equation = var_11722_equation_0, values = (var_11679_cast_fp16, var_11715_cast_fp16))[name = tensor("op_11722_cast_fp16")]; tensor var_11724_equation_0 = const()[name = tensor("op_11724_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11724_cast_fp16 = einsum(equation = var_11724_equation_0, values = (var_11683_cast_fp16, var_11716_cast_fp16))[name = tensor("op_11724_cast_fp16")]; tensor var_11726_equation_0 = const()[name = tensor("op_11726_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11726_cast_fp16 = einsum(equation = var_11726_equation_0, values = (var_11687_cast_fp16, var_11717_cast_fp16))[name = tensor("op_11726_cast_fp16")]; tensor var_11728_equation_0 = const()[name = tensor("op_11728_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11728_cast_fp16 = einsum(equation = var_11728_equation_0, values = (var_11691_cast_fp16, var_11718_cast_fp16))[name = tensor("op_11728_cast_fp16")]; tensor input_519_interleave_0 = const()[name = tensor("input_519_interleave_0"), val = tensor(false)]; tensor input_519_cast_fp16 = concat(axis = var_10509, interleave = input_519_interleave_0, values = (var_11720_cast_fp16, var_11722_cast_fp16, var_11724_cast_fp16, var_11726_cast_fp16, var_11728_cast_fp16))[name = tensor("input_519_cast_fp16")]; tensor var_11734 = const()[name = tensor("op_11734"), val = tensor([1, 1])]; tensor var_11736 = const()[name = tensor("op_11736"), val = tensor([1, 1])]; tensor var_11738_pad_type_0 = const()[name = tensor("op_11738_pad_type_0"), val = tensor("custom")]; tensor var_11738_pad_0 = const()[name = tensor("op_11738_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(835548480)))]; tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(835753344)))]; tensor var_11738_cast_fp16 = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_11736, groups = var_10509, pad = var_11738_pad_0, pad_type = var_11738_pad_type_0, strides = var_11734, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16, x = input_519_cast_fp16)[name = tensor("op_11738_cast_fp16")]; tensor inputs_cast_fp16 = add(x = var_11738_cast_fp16, y = inputs_93_cast_fp16)[name = tensor("inputs_cast_fp16")]; tensor var_11742 = const()[name = tensor("op_11742"), val = tensor([1])]; tensor channels_mean_cast_fp16 = reduce_mean(axes = var_11742, keep_dims = var_10504, x = inputs_cast_fp16)[name = tensor("channels_mean_cast_fp16")]; tensor zero_mean_cast_fp16 = sub(x = inputs_cast_fp16, y = channels_mean_cast_fp16)[name = tensor("zero_mean_cast_fp16")]; tensor zero_mean_sq_cast_fp16 = mul(x = zero_mean_cast_fp16, y = zero_mean_cast_fp16)[name = tensor("zero_mean_sq_cast_fp16")]; tensor var_11746 = const()[name = tensor("op_11746"), val = tensor([1])]; tensor var_11747_cast_fp16 = reduce_mean(axes = var_11746, keep_dims = var_10504, x = zero_mean_sq_cast_fp16)[name = tensor("op_11747_cast_fp16")]; tensor var_11748_to_fp16 = const()[name = tensor("op_11748_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_11749_cast_fp16 = add(x = var_11747_cast_fp16, y = var_11748_to_fp16)[name = tensor("op_11749_cast_fp16")]; tensor denom_epsilon_0_to_fp16 = const()[name = tensor("denom_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_cast_fp16 = rsqrt(epsilon = denom_epsilon_0_to_fp16, x = var_11749_cast_fp16)[name = tensor("denom_cast_fp16")]; tensor out_cast_fp16 = mul(x = zero_mean_cast_fp16, y = denom_cast_fp16)[name = tensor("out_cast_fp16")]; tensor var_11753_to_fp16 = const()[name = tensor("op_11753_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(835754048)))]; tensor var_11754_cast_fp16 = add(x = out_cast_fp16, y = var_11753_to_fp16)[name = tensor("op_11754_cast_fp16")]; tensor var_11756_to_fp16 = const()[name = tensor("op_11756_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(835754752)))]; tensor input_521_cast_fp16 = mul(x = var_11754_cast_fp16, y = var_11756_to_fp16)[name = tensor("input_521_cast_fp16")]; tensor var_11764 = const()[name = tensor("op_11764"), val = tensor([1, 1])]; tensor var_11766 = const()[name = tensor("op_11766"), val = tensor([1, 1])]; tensor var_11768_pad_type_0 = const()[name = tensor("op_11768_pad_type_0"), val = tensor("custom")]; tensor var_11768_pad_0 = const()[name = tensor("op_11768_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(835755456)))]; tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(837393920)))]; tensor var_11768_cast_fp16 = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16, dilations = var_11766, groups = var_10509, pad = var_11768_pad_0, pad_type = var_11768_pad_type_0, strides = var_11764, weight = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16, x = input_521_cast_fp16)[name = tensor("op_11768_cast_fp16")]; tensor var_11769_split_sizes_0 = const()[name = tensor("op_11769_split_sizes_0"), val = tensor([1280, 1280])]; tensor var_11769_axis_0 = const()[name = tensor("op_11769_axis_0"), val = tensor(1)]; tensor var_11769_cast_fp16_0, tensor var_11769_cast_fp16_1 = split(axis = var_11769_axis_0, split_sizes = var_11769_split_sizes_0, x = var_11768_cast_fp16)[name = tensor("op_11769_cast_fp16")]; tensor var_11771_mode_0 = const()[name = tensor("op_11771_mode_0"), val = tensor("EXACT")]; tensor var_11771_cast_fp16 = gelu(mode = var_11771_mode_0, x = var_11769_cast_fp16_1)[name = tensor("op_11771_cast_fp16")]; tensor input_523_cast_fp16 = mul(x = var_11769_cast_fp16_0, y = var_11771_cast_fp16)[name = tensor("input_523_cast_fp16")]; tensor var_11775 = const()[name = tensor("op_11775"), val = tensor([1, 1])]; tensor var_11777 = const()[name = tensor("op_11777"), val = tensor([1, 1])]; tensor var_11779_pad_type_0 = const()[name = tensor("op_11779_pad_type_0"), val = tensor("custom")]; tensor var_11779_pad_0 = const()[name = tensor("op_11779_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(837399104)))]; tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(838218368)))]; tensor var_11779_cast_fp16 = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_11777, groups = var_10509, pad = var_11779_pad_0, pad_type = var_11779_pad_type_0, strides = var_11775, weight = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16, x = input_523_cast_fp16)[name = tensor("op_11779_cast_fp16")]; tensor hidden_states_333_cast_fp16 = add(x = var_11779_cast_fp16, y = inputs_cast_fp16)[name = tensor("hidden_states_333_cast_fp16")]; tensor var_11781 = const()[name = tensor("op_11781"), val = tensor([2, 320, 48, 48])]; tensor input_525_cast_fp16 = reshape(shape = var_11781, x = hidden_states_333_cast_fp16)[name = tensor("input_525_cast_fp16")]; tensor var_11785 = const()[name = tensor("op_11785"), val = tensor([1, 1])]; tensor var_11787 = const()[name = tensor("op_11787"), val = tensor([1, 1])]; tensor hidden_states_pad_type_0 = const()[name = tensor("hidden_states_pad_type_0"), val = tensor("custom")]; tensor hidden_states_pad_0 = const()[name = tensor("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_blocks_3_attentions_2_proj_out_weight_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_proj_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(838219072)))]; tensor up_blocks_3_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(838423936)))]; tensor hidden_states_cast_fp16 = conv(bias = up_blocks_3_attentions_2_proj_out_bias_to_fp16, dilations = var_11787, groups = var_10509, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = var_11785, weight = up_blocks_3_attentions_2_proj_out_weight_to_fp16, x = input_525_cast_fp16)[name = tensor("hidden_states_cast_fp16")]; tensor input_527_cast_fp16 = add(x = hidden_states_cast_fp16, y = hidden_states_323_cast_fp16)[name = tensor("input_527_cast_fp16")]; tensor reshape_240_shape_0 = const()[name = tensor("reshape_240_shape_0"), val = tensor([2, 32, 10, 48, 48])]; tensor reshape_240_cast_fp16 = reshape(shape = reshape_240_shape_0, x = input_527_cast_fp16)[name = tensor("reshape_240_cast_fp16")]; tensor reduce_mean_180_axes_0 = const()[name = tensor("reduce_mean_180_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_180_keep_dims_0 = const()[name = tensor("reduce_mean_180_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_180_cast_fp16 = reduce_mean(axes = reduce_mean_180_axes_0, keep_dims = reduce_mean_180_keep_dims_0, x = reshape_240_cast_fp16)[name = tensor("reduce_mean_180_cast_fp16")]; tensor sub_120_cast_fp16 = sub(x = reshape_240_cast_fp16, y = reduce_mean_180_cast_fp16)[name = tensor("sub_120_cast_fp16")]; tensor square_60_cast_fp16 = square(x = sub_120_cast_fp16)[name = tensor("square_60_cast_fp16")]; tensor reduce_mean_182_axes_0 = const()[name = tensor("reduce_mean_182_axes_0"), val = tensor([2, 3, 4])]; tensor reduce_mean_182_keep_dims_0 = const()[name = tensor("reduce_mean_182_keep_dims_0"), val = tensor(true)]; tensor reduce_mean_182_cast_fp16 = reduce_mean(axes = reduce_mean_182_axes_0, keep_dims = reduce_mean_182_keep_dims_0, x = square_60_cast_fp16)[name = tensor("reduce_mean_182_cast_fp16")]; tensor add_120_y_0_to_fp16 = const()[name = tensor("add_120_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_120_cast_fp16 = add(x = reduce_mean_182_cast_fp16, y = add_120_y_0_to_fp16)[name = tensor("add_120_cast_fp16")]; tensor sqrt_60_cast_fp16 = sqrt(x = add_120_cast_fp16)[name = tensor("sqrt_60_cast_fp16")]; tensor real_div_60_cast_fp16 = real_div(x = sub_120_cast_fp16, y = sqrt_60_cast_fp16)[name = tensor("real_div_60_cast_fp16")]; tensor reshape_241_shape_0 = const()[name = tensor("reshape_241_shape_0"), val = tensor([2, 320, 48, 48])]; tensor reshape_241_cast_fp16 = reshape(shape = reshape_241_shape_0, x = real_div_60_cast_fp16)[name = tensor("reshape_241_cast_fp16")]; tensor add_121_gamma_0_to_fp16 = const()[name = tensor("add_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(838424640)))]; tensor add_121_beta_0_to_fp16 = const()[name = tensor("add_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(838425344)))]; tensor add_121_epsilon_0_to_fp16 = const()[name = tensor("add_121_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_121_cast_fp16 = batch_norm(beta = add_121_beta_0_to_fp16, epsilon = add_121_epsilon_0_to_fp16, gamma = add_121_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_241_cast_fp16)[name = tensor("add_121_cast_fp16")]; tensor input_cast_fp16 = silu(x = add_121_cast_fp16)[name = tensor("input_cast_fp16")]; tensor var_11801 = const()[name = tensor("op_11801"), val = tensor(1)]; tensor var_11804 = const()[name = tensor("op_11804"), val = tensor([1, 1])]; tensor var_11806 = const()[name = tensor("op_11806"), val = tensor([1, 1])]; tensor var_11808_pad_type_0 = const()[name = tensor("op_11808_pad_type_0"), val = tensor("custom")]; tensor var_11808_pad_0 = const()[name = tensor("op_11808_pad_0"), val = tensor([1, 1, 1, 1])]; tensor conv_out_weight_to_fp16 = const()[name = tensor("conv_out_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(838426048)))]; tensor conv_out_bias_to_fp16 = const()[name = tensor("conv_out_bias_to_fp16"), val = tensor([-0x1.4b4p-9, 0x1.6f4p-9, 0x1.9ap-12, 0x1.04p-9])]; tensor var_11808_cast_fp16 = conv(bias = conv_out_bias_to_fp16, dilations = var_11806, groups = var_11801, pad = var_11808_pad_0, pad_type = var_11808_pad_type_0, strides = var_11804, weight = conv_out_weight_to_fp16, x = input_cast_fp16)[name = tensor("op_11808_cast_fp16")]; tensor var_11808_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("op_11808_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; tensor noise_pred = cast(dtype = var_11808_cast_fp16_to_fp32_dtype_0, x = var_11808_cast_fp16)[name = tensor("cast_0")]; } -> (noise_pred); }