Skip to content

feat(vllm_batch): add reasoning_parser support for batch inference #804

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

Open
lilyz-ai wants to merge 2 commits into
base: main
Choose a base branch
from
+17 −2
Open

feat(vllm_batch): add reasoning_parser support for batch inference #804

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
926575f
feat(vllm_batch): add reasoning_parser support for batch inference
lilyz-ai Apr 3, 2026
0924eeb
fix(vllm_batch): filter serving-only args from AsyncEngineArgs
lilyz-ai Apr 3, 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
5 changes: 5 additions & 0 deletions model-engine/model_engine_server/common/dtos/llms/vllm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,6 +73,11 @@ class VLLMModelConfig(BaseModel):
description="Enable auto tool choice",
)

reasoning_parser: Optional[str] = Field(
None,
description="Reasoning parser (e.g. 'nemotron_v3', 'deepseek_r1')",
)

load_format: Optional[str] = Field(
None,
description="The format of the model weights to load.\n\n"
Expand Down
14 changes: 12 additions & 2 deletions model-engine/model_engine_server/inference/vllm/vllm_batch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -290,7 +290,15 @@ async def init_engine(
seed=request.model_cfg.seed or 0,
gpu_memory_utilization=request.max_gpu_memory_utilization or 0.9,
)
engine_args_dict = {**default_engine_args_dict, **parsed_configs.model_dump(exclude_none=True)}
_serving_only_keys = {"reasoning_parser"}
engine_args_dict = {
k: v
for k, v in {
**default_engine_args_dict,
**parsed_configs.model_dump(exclude_none=True),
}.items()
if k not in _serving_only_keys
}
Comment thread
greptile-apps[bot] marked this conversation as resolved.
Outdated
print("vLLM engine args:", engine_args_dict, flush=True)

engine_args = AsyncEngineArgs(**engine_args_dict)
Expand Down Expand Up @@ -319,7 +327,9 @@ async def init_engine(
return_tokens_as_token_ids=False,
enable_auto_tool_choice=False,
tool_call_parser=None,
structured_outputs_config=argparse.Namespace(reasoning_parser=None),
structured_outputs_config=argparse.Namespace(
reasoning_parser=parsed_configs.reasoning_parser
),
enable_prompt_tokens_details=False,
enable_force_include_usage=False,
enable_log_outputs=False,
Expand Down