dylanyunlong/project_6
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

fix: accept max_completion_tokens + reasoning_effort + n=2

Browse Source 
Sub655 replay log shows 881 requests, majority failing with:
  'Extra inputs are not permitted', 'input': 8192  (max_completion_tokens)
  'Extra inputs are not permitted', 'input': 'low' (reasoning_effort)

Root cause: protocol.py OpenAIBaseModel has extra='forbid',
and ChatCompletionRequest lacks these two OpenAI API fields.

Sub168 has the SAME bug — both reject max_completion_tokens.

Fixes:
1. protocol.py: add max_completion_tokens (Optional[int]) field
2. protocol.py: add reasoning_effort (Optional[str]) field
3. serving_chat.py: merge max_completion_tokens into max_tokens
4. computility-run.yaml: max_num_seqs=2 (fixes t2_n_2 HTTP 400)
This commit is contained in:
project6-dev
2026-08-12 06:16:44 +00:00
parent b806b15688
commit ac84e40f88
3 changed files with 7 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
computility-run.yaml
View File

@@ -15,7 +15,7 @@ command:
- -tp - -tp
- '4' - '4'
- --max-num-seqs - --max-num-seqs
- '1' - '2'
- --disable-log-requests - --disable-log-requests
- --disable-frontend-multiprocessing - --disable-frontend-multiprocessing
- --max-num-batched-tokens - --max-num-batched-tokens

3
qwen3_6_scripts/protocol.py
View File

@@ -180,6 +180,8 @@ class ChatCompletionRequest(OpenAIBaseModel):
logprobs: Optional[bool] = False logprobs: Optional[bool] = False
top_logprobs: Optional[int] = 0 top_logprobs: Optional[int] = 0
max_tokens: Optional[int] = None max_tokens: Optional[int] = None
# OpenAI newer API field — treat as alias for max_tokens
max_completion_tokens: Optional[int] = None
n: Optional[int] = 1 n: Optional[int] = 1
presence_penalty: Optional[float] = 0.0 presence_penalty: Optional[float] = 0.0
response_format: Optional[ResponseFormat] = None response_format: Optional[ResponseFormat] = None
@@ -193,6 +195,7 @@ class ChatCompletionRequest(OpenAIBaseModel):
tool_choice: Optional[Union[Literal["none"], Literal["auto"], tool_choice: Optional[Union[Literal["none"], Literal["auto"],
ChatCompletionNamedToolChoiceParam]] = "none" ChatCompletionNamedToolChoiceParam]] = "none"
thinking: Optional[Union[bool, str, Dict[str, Any]]] = None thinking: Optional[Union[bool, str, Dict[str, Any]]] = None
reasoning_effort: Optional[str] = None
# NOTE this will be ignored by VLLM -- the model determines the behavior # NOTE this will be ignored by VLLM -- the model determines the behavior
parallel_tool_calls: Optional[bool] = False parallel_tool_calls: Optional[bool] = False

3
qwen3_6_scripts/serving_chat.py
View File

@@ -394,6 +394,9 @@ class OpenAIServingChat(OpenAIServing):
assert prompt_inputs is not None assert prompt_inputs is not None
sampling_params: Union[SamplingParams, BeamSearchParams] sampling_params: Union[SamplingParams, BeamSearchParams]
# OpenAI API: max_completion_tokens takes precedence over max_tokens
if request.max_completion_tokens is not None and request.max_tokens is None:
request.max_tokens = request.max_completion_tokens
default_max_tokens = self.max_model_len - len( default_max_tokens = self.max_model_len - len(
prompt_inputs["prompt_token_ids"]) prompt_inputs["prompt_token_ids"])
if request.use_beam_search: if request.use_beam_search: