On GPU, always expand to beam size before the first decoding step #263
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Outdated suggestions cannot be applied.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.
This is a partial revert of 7ebba2e which seems to cause a performance regression when using large beam size on GPU.
The performance regression can be observed on both FP32 and FP16 decoding, but not with the same beam size. This could indicate that the issue is related to the caching allocator that does not like the larger batch size at the second step.
The behavior is unchanged on CPU.