Skip to content

[Bug] Qwen reasoning parser #27241

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

Closed
wants to merge 6 commits into from
Closed

[Bug] Qwen reasoning parser #27241

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
2ced789
allow runai_streamer_sharded in config check
ahao-anyscale Oct 15, 2025
6ba712c
Merge branch 'vllm-project:main' into main
ahao-anyscale Oct 21, 2025
553817e
added override for reasoning parser for qwen3
ahao-anyscale Oct 21, 2025
98ab1cd
Merge remote-tracking branch 'origin/main' into qwen-reasoning-parser
ahao-anyscale Oct 21, 2025
2db5c63
Merge branch 'main' into qwen-reasoning-parser
ahao-anyscale Oct 22, 2025
70d1cdb
Merge branch 'main' into qwen-reasoning-parser
ahao-anyscale Oct 24, 2025
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
9 changes: 9 additions & 0 deletions vllm/model_executor/models/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -489,6 +489,14 @@ def verify_and_update_config(cls, vllm_config: "VllmConfig") -> None:
logger.info("Using bfloat16 kv-cache for DeepSeekV3.2")


class Qwen3MoeForCausalLMConfig(VerifyAndUpdateConfig):
@staticmethod
def verify_and_update_config(vllm_config: "VllmConfig") -> None:
structured_outputs_config = vllm_config.structured_outputs_config
if structured_outputs_config.reasoning_parser == "":
Copy link
Collaborator

@chaunceyjiang chaunceyjiang Oct 25, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How do you start vLLM?

This field structured_outputs_config.reasoning_parser can be set via the parameter --reasoning-parser qwen3.

Copy link
Contributor Author

@ahao-anyscale ahao-anyscale Oct 27, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is the user intended to supply these arguments? I seems that it would make more sense to automatically set them for reasoning models. I saw that gptoss had its reasoning-parser set automatically via verify_and_update_config. Thanks for reviewing!

Copy link
Member

@hmellor hmellor Oct 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes I think right now it is expected that the user opts in to reasoning https://docs.vllm.ai/en/latest/features/reasoning_outputs.html

structured_outputs_config.reasoning_parser = "qwen3"


MODELS_CONFIG_MAP: dict[str, type[VerifyAndUpdateConfig]] = {
"GteModel": SnowflakeGteNewModelConfig,
"GteNewModel": GteNewModelConfig,
Expand All @@ -498,6 +506,7 @@ def verify_and_update_config(cls, vllm_config: "VllmConfig") -> None:
"Qwen2ForProcessRewardModel": Qwen2ForProcessRewardModelConfig,
"Qwen2ForRewardModel": Qwen2ForRewardModelConfig,
"Qwen3ForSequenceClassification": Qwen3ForSequenceClassificationConfig,
"Qwen3MoeForCausalLM": Qwen3MoeForCausalLMConfig,
"XLMRobertaModel": JinaRobertaModelConfig,
"JinaVLForRanking": JinaVLForSequenceClassificationConfig,
"JambaForSequenceClassification": JambaForSequenceClassificationConfig,
Expand Down