Sync transformers and accelerate versions#562
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
dacorvo
left a comment
dacorvo
left a comment
There was a problem hiding this comment.
Awesome ! Looks good to me, but found a few nits !
| tr_loss_div = tr_loss / dp_size | ||
| tr_loss_div = tr_loss / dp_size | ||
|
|
||
| xm.mark_step() |
There was a problem hiding this comment.
I love when these lines pop magically to solve a sync issue ... 😜
| ignore_keys_for_eval=ignore_keys_for_eval, | ||
| **kwargs, | ||
| ) | ||
| # with hub_neuronx_cache("training", entry=self.model_cache_entry): |
There was a problem hiding this comment.
Are you sure you don't want to fetch from the cache here ? If so, you should remove the commented line.
There was a problem hiding this comment.
Fixed, it was a quick test artifact
|
You should unpin the safetensor package here because there is now a conflict: |
…te_dacorvo' into sync_transformers_and_accelerate
dacorvo
left a comment
dacorvo
left a comment
There was a problem hiding this comment.
LGTM, thanks ! If you can fix the seq2seq tracing issue before merging that pull-request that is even better.
|
I fixed all but one test:
|
3 participants
What does this PR do?
This PR synchronizes
optimum-neuronwith more recenttransformersandaccelerateversions:accelerate==0.29.2, which is the latest release when this PR is being done,transformers==4.40.2, which will be the latest releae when this PR is merged.Related PR in
transformers: huggingface/transformers#30259On top of that: