feat: support cached cos/sin in rope APIs #585
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Closed
# for free
to join this conversation on GitHub.
Already have an account?
# to comment
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Outdated suggestions cannot be applied.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.
As requested in #530 , this PR implements the RoPE with cached cos/sin embeddings, which is more flexible in some use cases.
In our previous RoPE implementations, cos/sin values are computed on-the-fly inside kernels with float32 instead using cached values.
In this PR we found that if we use f16 cos/sin cache, the rope result will have a large discrepancy compared to our original implementation
flashinfer.apply_rope(which stores cos/sin with fp32). So we require thecos_cacheandsin_cacheto use fp32 data type.cc @dreaming-panda @ByronHsu