| { | |
| "_name_or_path": "SEWY3-math-r0", | |
| "architectures": [ | |
| "Sewy3ForCausalLM" | |
| ], | |
| "attention_bias": false, | |
| "attention_dropout": 0.0, | |
| "attn_logit_softcapping": 50.0, | |
| "bos_token_id": 1, | |
| "cache_implementation": "hybrid", | |
| "eos_token_id": 2, | |
| "final_logit_softcapping": 30.0, | |
| "head_dim": 256, | |
| "hidden_act": "silu", | |
| "hidden_activation": "silu", | |
| "hidden_size": 400, | |
| "initializer_range": 0.02, | |
| "intermediate_size": 1600, | |
| "max_position_embeddings": 4096, | |
| "model_type": "gemma", | |
| "num_attention_heads": 16, | |
| "num_hidden_layers": 28, | |
| "num_key_value_heads": 4, | |
| "num_ve": 3, | |
| "pad_token_id": 0, | |
| "query_pre_attn_scalar": 256, | |
| "rms_norm_eps": 1e-06, | |
| "rope_theta": 1000000.0, | |
| "sliding_window": 512, | |
| "tie_word_embeddings": false, | |
| "torch_dtype": "bfloat16", | |
| "transformers_version": "4.48.1", | |
| "use_cache": true, | |
| "use_sliding_window": true, | |
| "vocab_size": 49152 | |
| } | |