[ROCm][Bugfix] Fix Triton W4A16 bug in determining if transpose is required for GPTQ/AutoGPTQ - #48998
Merged
Conversation
… layout (vllm-project#47770)" This commit breaks asymm quantized models like cyankiwi-MiniMax-M3-AWQ-INT4. Signed-off-by: Qiang Li <qiang.li2@amd.com>
…quired for GPTQ/AutoGPTQ Signed-off-by: Qiang Li <qiang.li2@amd.com>
qli88
requested review from
AndreasKaratzas,
WoosukKwon,
mgoin,
tlrmchlsmth,
yewentao256 and
zyongye
as code owners
July 18, 2026 00:07
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
1 task
BowenBao
approved these changes
Jul 21, 2026
BowenBao
left a comment
BowenBao
left a comment
Contributor
There was a problem hiding this comment.
Fix LGTM, minor comment.
AndreasKaratzas
added
the
ready
Member
|
/ci run |
|
✅ Triggered Buildkite CI #83777 for commit |
tjtanaa
approved these changes
Aug 17, 2026
Member
|
/ci run |
tjtanaa
enabled auto-merge (squash)
August 17, 2026 11:01
|
✅ Triggered Buildkite CI #84195 for commit |
6 tasks
zyp2014
pushed a commit
to zyp2014/vllm
that referenced
this pull request
Aug 21, 2026
…quired for GPTQ/AutoGPTQ (vllm-project#48998) Signed-off-by: Qiang Li <qiang.li2@amd.com>
wyettzeng
pushed a commit
to wyettzeng/vllm
that referenced
this pull request
Aug 21, 2026
…quired for GPTQ/AutoGPTQ (vllm-project#48998) Signed-off-by: Qiang Li <qiang.li2@amd.com> Signed-off-by: Wyett <wyettzeng@gmail.com>
zufangzhu
pushed a commit
to zufangzhu/vllm
that referenced
this pull request
Aug 24, 2026
…quired for GPTQ/AutoGPTQ (vllm-project#48998) Signed-off-by: Qiang Li <qiang.li2@amd.com> Signed-off-by: Zhu, Zufang <zufang.zhu@intel.com>
khushali9
pushed a commit
to khushali9/vllm
that referenced
this pull request
Aug 29, 2026
…quired for GPTQ/AutoGPTQ (vllm-project#48998) Signed-off-by: Qiang Li <qiang.li2@amd.com> Signed-off-by: khushali9 <khushali.desai9@gmail.com>
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
4 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.
Purpose
this commit #47770 (for issue #47159) introduced a shape-based method to determine if qzeros need transposing, but the shape check is ambiguous when the two candidate shapes are identical. (cyankiwi-MiniMax-M3-AWQ-INT4 with tp=8 as an example). This PR is to fix this issue:
Test Plan
raydelossantos/Qwen3.6-27B-GPTQ-Int4:
vllm serve /data/raydelossantos-Qwen3.6-27B-GPTQ-Int4/
--attention-backend TRITON_ATTN
--max-model-len 65536
--reasoning-parser qwen3
--enable-auto-tool-choice \
--tool-call-parser qwen3_coder
cyankiwi/MiniMax-M3-AWQ-INT4:
VLLM_USE_BREAKABLE_CUDAGRAPH=0 vllm serve /data/cyankiwi-MiniMax-M3-AWQ-INT4/
--block-size 128
--tensor-parallel-size 8
--attention-backend TRITON_ATTN
--tool-call-parser minimax_m3
--enable-auto-tool-choice
--reasoning-parser minimax_m3
--moe-backend emulation
Both work as expected.