Conversation
yizhang2077
requested review from
BBuf,
Edwardf0t1,
HaiShaw,
Ying1123,
ch-wan,
ispobock,
kushanam,
merrymercy and
zhyncs
as code owners
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Summary of Changes
Hello @yizhang2077, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!
This pull request resolves a numerical accuracy issue in the chunk_scaled_dot_kkt_fwd_kernel by reordering a critical scaling operation. The b_beta scaling, previously applied prematurely, is now correctly applied to the b_A matrix after its initial computation and exponential adjustment. This change is crucial for preventing incorrect mathematical accumulation, particularly in H20, and ensures the kernel's calculations maintain numerical stability and precision.
Highlights
- Numerical Stability Fix: Addressed an Incorrect Mathematical Accumulation (IMA) behavior within the
chunk_scaled_dot_kkt_fwd_kernelfunction, specifically impacting H20 environments. - Reordered Scaling Operation: The
b_betascaling factor is now applied to the accumulatedb_Amatrix after the dot product andsafe_expcalculation, rather than pre-scaling theb_kvector. - Kernel Logic Adjustment: Modified the order of operations in the
chunk_scaled_dot_kkt_fwd_kernelto ensurekktscaling occurs post-reduction, aligning with the intended mathematical sequence.
Using Gemini Code Assist
The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.
Invoking Gemini
You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in pull request comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
Customization
To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.
Limitations & Feedback
Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.
You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.
Footnotes
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
There was a problem hiding this comment.
Code Review
This pull request refactors the chunk_scaled_dot_kkt_fwd_kernel Triton kernel to improve performance and address a potential hardware-specific issue. The change moves the scaling by beta out of the accumulation loop. This is a sound optimization that is mathematically equivalent to the previous implementation but more efficient, as it reduces redundant computations within the loop. This change aligns with a fix from the upstream flash-linear-attention repository, which reportedly resolves an in-memory accumulation (IMA) issue on NVIDIA's Hopper architecture. The changes are correct and beneficial, and I have no further suggestions.
* origin/qwen3: (30 commits) chore: bump sgl-kernel 0.3.11 (sgl-project#10630) feat: add fused moe config for Qwen3-Next-80B-A3B-Instruct on B200 (sgl-project#10631) model support: Sarashina2VisionForCausalLM (sgl-project#10632) [Performance] Qwen3-Next: speed up update_mamba_state_after_mtp_verify by 10x; e2e up to 3.54% faster (sgl-project#10586) [Performance] Qwen3-Next: replace arange to cached query_start_loc_li… (sgl-project#10553) [Feature] Speculative decoding support lookahead (sgl-project#9873) refactor: use registry for _get_attention_backend_from_str (sgl-project#10629) [router] refactor worker to builder pattern 1/n (sgl-project#10628) Garbage collector regression in the online server (sgl-project#10621) feat: Add FlexAttention Backend for Efficient Sparse Attention (sgl-project#9947) Fix bias handling in TritonMoeQuantInfo within quantization/mxfp4.py (sgl-project#10579) [Performance] qwen3-next improve causal conv1d in prefill phase (sgl-project#10595) Fix sgl_kernel import failure on devices other than CUDA (sgl-project#10610) support qwen3-next-fp8 deepep (sgl-project#10622) update deepep version for qwen3-next deepep moe (sgl-project#10624) Feat/add heartbeat mechanism for nixl conn (sgl-project#10222) [RL] Add destroy process group api (sgl-project#9979) fix deepep assert when PD disaggregation == null (sgl-project#8274) Scale kkt after reduction (sgl-project#10604) [improvement] add average input/output token length for hicache benchmark stats output (sgl-project#10525) ...
|
Hello, I want to know why it works, and whether the bug only occurs in H20 or all the GPU？Is there some difference between scaling before reduction and after reduction？ |
4 participants
Motivation
ref link fla-org/flash-linear-attention@d57a3b0, this pr could fix
chunk_scaled_dot_kkt_fwd_kernelIMA behavior in H20Modifications
Accuracy Tests
Benchmarking and Profiling
Checklist