From 375e5984fec8f79f1ec4190c2fd76cc185f6a58f Mon Sep 17 00:00:00 2001 From: Shiyan Deng Date: Wed, 14 Jan 2026 20:07:26 -0800 Subject: [PATCH] Support configure skip_special_tokens in openai response api (#32345) Signed-off-by: Shiyan Deng --- vllm/entrypoints/openai/responses/protocol.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/vllm/entrypoints/openai/responses/protocol.py b/vllm/entrypoints/openai/responses/protocol.py index 0ff67373a..4a4c32cc6 100644 --- a/vllm/entrypoints/openai/responses/protocol.py +++ b/vllm/entrypoints/openai/responses/protocol.py @@ -171,6 +171,7 @@ class ResponsesRequest(OpenAIBaseModel): top_k: int | None = None truncation: Literal["auto", "disabled"] | None = "disabled" user: str | None = None + skip_special_tokens: bool = True # --8<-- [start:responses-extra-params] request_id: str = Field( @@ -279,6 +280,7 @@ class ResponsesRequest(OpenAIBaseModel): structured_outputs=structured_outputs, logit_bias=self.logit_bias, skip_clone=True, # Created fresh per request, safe to skip clone + skip_special_tokens=self.skip_special_tokens, ) def is_include_output_logprobs(self) -> bool: