feat: support huggingface transformer style rope interface #568
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
yzh119
pushed a commit
that referenced
this pull request
Oct 30, 2024
Fix after changes made in #568 torch.compile doesn't like returning input arguments. So, change the return type of pybind fns to `void`, given that it's already an inplace op. PyTorch Library annotation is applied to a wrapper function for each pybind op. Python API doesn't change. Both inplace and non-inplace versions calls the annotated wrapper function.
tsu-bin
added a commit
to tsu-bin/flashinfer_dev
that referenced
this pull request
Nov 5, 2024
tsu-bin
added a commit
to tsu-bin/flashinfer_dev
that referenced
this pull request
Nov 5, 2024
yzh119
pushed a commit
that referenced
this pull request
Nov 5, 2024
Hi, now tvm wrapper build failed cause by #568. I noticed that the new `BatchQKApplyRotary` interface removed `__restrict__` modifier from `DType* q, DType* k, DType* q_rope, DType* k_rope`, so it's trivial to just add one adapter function to fix this issue. Co-authored-by: tsu-bin <[email protected]>
Closed
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in 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Outdated suggestions cannot be applied.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.
Previously our rope apis assume the position indices of each request is contiguous, which is not appropriate for applications such as speculative decoding, this PR fixes the issue by supporting the huggingface transformer-style API which use
pos_idsargument to specify positions.This PR implements parts of the feature of #530 , other requests are coming in later PRs.
cc @dreaming-panda @abcdabcd987 @ByronHsu