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 2, 11:00 UTC
[Frontend]Responses API supports chat_template_kwargs (#43761)
Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
Chauncey committed
May 29, 2026 at 15:58 UTC
87f12e5c7c8eed62e295a63f1a12ed438a45de1a
1 file changed
+13
-2
vllm/entrypoints/openai/responses/serving.py
+13
-2
@@ -1041,7 +1041,10 @@ class OpenAIServingResponses(OpenAIServing):
1041
1042
# Use parser to extract and create response output items
1043
if self.parser:
1044
- parser = self.parser(tokenizer, request.tools)
1044
+ chat_template_kwargs = self._effective_chat_template_kwargs(request)
1045
+ parser = self.parser(
1046
+ tokenizer, request.tools, chat_template_kwargs=chat_template_kwargs
1047
+ )
1048
return parser.extract_response_outputs(
1049
model_output=final_output.text,
1050
model_output_token_ids=final_output.token_ids,
@@ -1378,7 +1381,15 @@ class OpenAIServingResponses(OpenAIServing):
1381
],
1382
) -> AsyncGenerator[StreamingResponsesResponse, None]:
1383
processor = SimpleStreamingEventProcessor()
1381
- parser = self.parser(tokenizer, request.tools) if self.parser else None
1384
+ parser = (
1385
+ self.parser(
1386
+ tokenizer,
1387
+ request.tools,
1388
+ chat_template_kwargs=self._effective_chat_template_kwargs(request),
1389
+ )
1390
+ if self.parser
1391
+ else None
1392
+ )
1393
1394
def _get_logprobs(
1395
output: CompletionOutput,