Skip device placement for past key values in decoder models #23919
+19
−1
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
This PR skips the device placement for the
past_key_values
in big models, which is responsible for a lot of time lost according to the analysis in this issue. The idea is that the past key values (one per layer) are all generated on the device of the layer they correspond to and never need to be moved.Needs to be tested with Accelerate at huggingface/accelerate#1491 (until this PR is merged).