Commit
•
553cf70
1
Parent(s):
a709b51
fix mixed precision loading with recent transformers versions
Browse files- modeling_xlm_roberta.py +1 -0
modeling_xlm_roberta.py
CHANGED
|
@@ -404,6 +404,7 @@ class XLMRobertaPreTrainedModel(PreTrainedModel):
|
|
| 404 |
config_class = XLMRobertaFlashConfig
|
| 405 |
base_model_prefix = "roberta"
|
| 406 |
supports_gradient_checkpointing = True
|
|
|
|
| 407 |
|
| 408 |
def _set_gradient_checkpointing(self, module, value=False):
|
| 409 |
if isinstance(module, XLMRobertaEncoder):
|
|
|
|
| 404 |
config_class = XLMRobertaFlashConfig
|
| 405 |
base_model_prefix = "roberta"
|
| 406 |
supports_gradient_checkpointing = True
|
| 407 |
+
_supports_param_buffer_assignment = False
|
| 408 |
|
| 409 |
def _set_gradient_checkpointing(self, module, value=False):
|
| 410 |
if isinstance(module, XLMRobertaEncoder):
|