[Feature] limit thinking tokens (hard limit) (#20859)

Signed-off-by: Sungjae Lee <33976427+llsj14@users.noreply.github.com>
Signed-off-by: Sungjae Lee <sung-jae.lee@navercorp.com>
Signed-off-by: Chauncey <chaunceyjiang@gmail.com>
Co-authored-by: Chauncey <chaunceyjiang@gmail.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Sungjae Lee
2026-03-24 09:53:07 -07:00
committed by GitHub
co-authored by Chauncey Harry Mellor
parent 8de5261e69
commit 4731884796
13 changed files with 702 additions and 12 deletions
@@ -180,6 +180,7 @@ class ChatCompletionRequest(OpenAIBaseModel):
| None
) = "none"
reasoning_effort: Literal["none", "low", "medium", "high"] | None = None
thinking_token_budget: int | None = None
include_reasoning: bool = True
parallel_tool_calls: bool | None = True
@@ -515,6 +516,7 @@ class ChatCompletionRequest(OpenAIBaseModel):
structured_outputs=self.structured_outputs,
logit_bias=self.logit_bias,
bad_words=self.bad_words,
thinking_token_budget=self.thinking_token_budget,
allowed_token_ids=self.allowed_token_ids,
extra_args=extra_args or None,
skip_clone=True, # Created fresh per request, safe to skip clone