[cuTile] Add rope/qwen2vl_mrope/kl_div/group_norm/multi_token_attention#1269
Open
xjmxyt wants to merge 1 commit into
Open
[cuTile] Add rope/qwen2vl_mrope/kl_div/group_norm/multi_token_attention#1269xjmxyt wants to merge 1 commit into
xjmxyt wants to merge 1 commit into
Conversation
cuTile implementations for group_norm, kl_div, llama4_rope, qwen2vl_mrope, rope, sparsemax, tiled_mlp, and multi_token_attention, dispatched via LIGER_KERNEL_IMPL=cutile. Includes three-way (liger_triton / liger_cutile / torch|huggingface) speed+memory benchmark data on B200 generated with the open-source pip nvidia-cuda-tileiras 13.3.36. Notable kernel choices: - rope/qwen2vl_mrope: stay in input dtype, drop redundant .contiguous() copies - kl_div: drop scale constexpr to avoid per-iter JIT recompile - group_norm: fp32 stats for numerical parity - multi_token_attention: conv-backward runs under the same cuDNN heuristic as the Triton path (no forced cudnn.benchmark) for apples-to-apples comparison Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
xjmxyt
force-pushed
the
jinmanx/add_kernel_v2
branch
from
13e5472 to
5fb9475
Compare
Contributor
Author
|
@vaibhavjindal @Tcc0403 Could you please help to review this? Thanks a lot. |
Tcc0403
approved these changes
Jul 8, 2026
Tcc0403
left a comment
Tcc0403
left a comment
Collaborator
There was a problem hiding this comment.
LGTM, @vaibhavjindal for a second look
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
cuTile implementations for group_norm, kl_div, llama4_rope, qwen2vl_mrope, rope, sparsemax, tiled_mlp, and multi_token_attention, dispatched via LIGER_KERNEL_IMPL=cutile. Includes three-way (liger_triton / liger_cutile / torch|huggingface) speed+memory benchmark data on B200 generated with the open-source pip nvidia-cuda-tileiras 13.3.36.
Summary
Testing Done
make testto ensure correctnessmake checkstyleto ensure code stylemake test-convergenceto ensure convergence