A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 06:00 UTC
[Misc][Frontend] passthrough `bad_words` (#19564)
Signed-off-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Co-authored-by: Francesco Bertolotti <francesco.bertolotti@igenius.ai> Co-authored-by: Aaron Pham <Aaronpham0103@gmail.com>
Francesco Bertolotti committed
Jun 16, 2025 at 07:05 UTC
367871a469202bc671b7fb39762d9ad86a839f7e
1 file changed
+2
vllm/entrypoints/openai/protocol.py
+2
@@ -272,6 +272,7 @@ class ChatCompletionRequest(OpenAIBaseModel):
272
truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]] = None
273
prompt_logprobs: Optional[int] = None
274
allowed_token_ids: Optional[list[int]] = None
275
+ bad_words: list[str] = Field(default_factory=list)
276
# --8<-- [end:chat-completion-sampling-params]
277
278
# --8<-- [start:chat-completion-extra-params]
@@ -550,6 +551,7 @@ class ChatCompletionRequest(OpenAIBaseModel):
551
else RequestOutputKind.FINAL_ONLY,
552
guided_decoding=guided_decoding,
553
logit_bias=self.logit_bias,
554
+ bad_words= self.bad_words,
555
allowed_token_ids=self.allowed_token_ids,
556
extra_args=({"kv_transfer_params": self.kv_transfer_params}
557
if self.kv_transfer_params else None))