Skip to content

Fix npu qwen3moe grpo #118

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
tastelikefeet merged 5 commits into from
Mar 30, 2026
Merged

Fix npu qwen3moe grpo #118

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
0702fc0
fix: patch moe weight loader before base weight sync
vx120 Mar 18, 2026
1f44b32
Merge branch 'modelscope:main' into fix_npu_moe_grpo
vx120 Mar 18, 2026
6f4e9ed
fix: apply pre-commit changes
vx120 Mar 18, 2026
c2f520e
Merge branch 'modelscope:main' into fix_npu_moe_grpo
vx120 Mar 30, 2026
2860e2e
patch qwen3moe by existed code
vx120 Mar 30, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 1 addition & 2 deletions src/twinkle/sampler/vllm_sampler/vllm_sampler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -122,8 +122,7 @@ def __init__(self, model_id: str, engine_args: Dict[str, Any] = None, device_mes
# fix: On NPU, monkey_patch_model can trigger Triton compatibility errors and abort sampler init.
# fix: Explicitly skip this patch on NPU and keep it for non-NPU paths only.
# NPU platform may trigger triton errors with monkey_patch_model
if Platform.get_platform().device_prefix() != 'npu':
self._run_in_loop(self.engine.engine.collective_rpc('monkey_patch_model'))
self._run_in_loop(self.engine.engine.collective_rpc('monkey_patch_model'))

VLLMLoraWeights()(self)

Expand Down
Loading