remove import/call to code no longer in latest transformers
#3
by
winglian
- opened
- modeling_deepseek.py +1 -7
modeling_deepseek.py
CHANGED
@@ -42,10 +42,7 @@ from transformers.modeling_outputs import (
|
|
42 |
SequenceClassifierOutputWithPast,
|
43 |
)
|
44 |
from transformers.modeling_utils import PreTrainedModel
|
45 |
-
from transformers.pytorch_utils import
|
46 |
-
ALL_LAYERNORM_LAYERS,
|
47 |
-
is_torch_greater_or_equal_than_1_13,
|
48 |
-
)
|
49 |
from transformers.utils import (
|
50 |
add_start_docstrings,
|
51 |
add_start_docstrings_to_model_forward,
|
@@ -67,9 +64,6 @@ if is_flash_attn_2_available():
|
|
67 |
# This makes `_prepare_4d_causal_attention_mask` a leaf function in the FX graph.
|
68 |
# It means that the function will not be traced through and simply appear as a node in the graph.
|
69 |
if is_torch_fx_available():
|
70 |
-
if not is_torch_greater_or_equal_than_1_13:
|
71 |
-
import torch.fx
|
72 |
-
|
73 |
_prepare_4d_causal_attention_mask = torch.fx.wrap(_prepare_4d_causal_attention_mask)
|
74 |
|
75 |
|
|
|
42 |
SequenceClassifierOutputWithPast,
|
43 |
)
|
44 |
from transformers.modeling_utils import PreTrainedModel
|
45 |
+
from transformers.pytorch_utils import ALL_LAYERNORM_LAYERS
|
|
|
|
|
|
|
46 |
from transformers.utils import (
|
47 |
add_start_docstrings,
|
48 |
add_start_docstrings_to_model_forward,
|
|
|
64 |
# This makes `_prepare_4d_causal_attention_mask` a leaf function in the FX graph.
|
65 |
# It means that the function will not be traced through and simply appear as a node in the graph.
|
66 |
if is_torch_fx_available():
|
|
|
|
|
|
|
67 |
_prepare_4d_causal_attention_mask = torch.fx.wrap(_prepare_4d_causal_attention_mask)
|
68 |
|
69 |
|