Revert "[CI][XPU] Disable model runner V2 for XPU quantization test for some partially pre-quantized models" - #56394
Merged
Conversation
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
mergify
Bot
added
ci/build
quantization
intel-gpu
…or some partially pre-quantized models" (vllm-project#56179) This reverts the workaround added in vllm-project#56179 that disabled model runner V2 (VLLM_USE_V2_MODEL_RUNNER=0) for the XPU quantization CI job. Signed-off-by: Chaojun Zhang <chaojun.zhang@intel.com>
chaojun-zhang
force-pushed
the
revert-56179-main
branch
from
September 11, 2026 04:49
4aa86b6 to
a7c72c7
Compare
chaojun-zhang
marked this pull request as draft
September 11, 2026 04:50
chaojun-zhang
changed the title
chaojun-zhang
changed the title
3 tasks
chaojun-zhang
changed the title
chaojun-zhang
marked this pull request as ready for review
September 11, 2026 12:10
jikunshang
approved these changes
Sep 14, 2026
Member
|
/ci run |
|
✅ Triggered Buildkite CI #88697 for commit |
Shreya-gaur
pushed a commit
to Shreya-gaur/vllm_private
that referenced
this pull request
Sep 14, 2026
…or some partially pre-quantized models" (vllm-project#56394) Signed-off-by: Chaojun Zhang <chaojun.zhang@intel.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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This reverts #56179.
The problem
test_online_quantization_loads_real_weights[...tinysmokeqwen3moe-W4A16...]hangs on XPU CI.CompressedTensorsWNA16MoEMethod.create_weights()(pre-#54809) always registers 4 g_idx-relatednn.Parameters, even for checkpoints (like this one) with no actorder/g_idx data — leaving them uninitialized.Verified on real XPU hardware
Re-registering just these 4 parameters reproduces the hang; removing them fixes it. On XPU,
_process_weights_xpu()never touches g_idx at all, so these parameters stay untouched throughout loading.Why not on CUDA
On CUDA (Marlin backend),
process_weights_after_loading()reads and replaces the g_idx parameter even for non-actorder checkpoints, so it doesn't stay untouched the way it does on XPU.Fixing PR
#54809 removes GPTQ actorder support and, as part of that, deletes this unconditional registration.
Why this revert
#54809 already fixes this on
main, so #56179'sVLLM_USE_V2_MODEL_RUNNER=0workaround is no longer needed.