fix: model.forward now accepts return_dict via kwargs #443
Conversation
|
@JamesKunstle please take a look. I am not sure about the change since I don't grasp the full extent of the override; I think I understand the root cause and I did my best to clean the code up from mere mechanical code reading. Someone with proper expertise would need to check if this makes in a broader context of what the code is trying to achieve beyond... not crashing. |
Since huggingface/transformers#36794 it handles the return_dict with can_return_tuple decorator that expects the argument to be passed as kwarg. Without the patch, instructlab functional tests fail with: AttributeError: 'bool' object has no attribute 'unsqueeze' This is because the passed return_dict argument is incorrectly interpreted as cache_position. Of course, the way the model forward method is overridden here is quite problematic and would benefit from a refactor. This patch is a stop-gap to fix the CI with minimal changes. Signed-off-by: Ihar Hrachyshka <[email protected]>
booxter
changed the title
|
e2e fails with |
|
Apparently e2e failure was supposed to be fixed by instructlab/instructlab#3254 but since this PR fixes functional tests for that other PR, we have chicken and egg. I guess instructlab will have to pin / cap transformers library first; then we can later land this change here; then finally bump in instructlab and uncap transformers. |
3 participants
Since huggingface/transformers#36794 it handles
the return_dict with can_return_tuple decorator that expects the
argument to be passed as kwarg.
Without the patch, instructlab functional tests fail with:
AttributeError: 'bool' object has no attribute 'unsqueeze'
This is because the passed return_dict argument is incorrectly
interpreted as cache_position.
Of course, the way the model forward method is overridden here is quite
problematic and would benefit from a refactor. This patch is a stop-gap
to fix the CI with minimal changes.
Signed-off-by: Ihar Hrachyshka [email protected]