fix(rocm): propagate hw_kernel_config to qwen35 layers#970
Open
chengshu-lcc wants to merge 1 commit intoalibaba:mainfrom
Open
fix(rocm): propagate hw_kernel_config to qwen35 layers#970chengshu-lcc wants to merge 1 commit intoalibaba:mainfrom
chengshu-lcc wants to merge 1 commit intoalibaba:mainfrom
Conversation
Collaborator
AI Code Review - PR #970Status: BLOCKING Summary: P0/0 · P1/1 · P2/0 · P3/0 Blocking IssuesP1
Checklist Violations (4 fail / 63 total)General Principles Checklist
RTP-LLM Checklist
Strengths
|
a98afd6 to
4ccec20
Compare
Collaborator
AI Code Review - PR #970Status: LGTM Summary: P0/0 · P1/0 · P2/1 · P3/0 lgtm ready to ci Non-blocking SuggestionsP2
Checklist ✅ (74 items passed)Strengths
|
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
概述
修复 ROCm 上 Qwen3-Next 在
USE_SWIZZLEA=1时无法走通 swizzled-A hipBLASLt 路径导致精度发散的问题。RocmImpl的 swizzle-A 权重列表中补上W.linear_attn_ba_w,让in_proj_ba与其他 linear attention 权重一起完成预 swizzle。hw_kernel_config从Qwen3NextModel一路透传到Qwen3NextDecoderLayer→Qwen3NextAttention/Qwen3NextGatedDeltaNet/DenseMLP,以及GenericMoeLayer的 shared expert,使LinearFactory.create_linear_from_weights能正确选中 swizzled-A kernel。