diff --git a/vllm/v1/engine/output_processor.py b/vllm/v1/engine/output_processor.py index de94a0e5d0..dc572ccc10 100644 --- a/vllm/v1/engine/output_processor.py +++ b/vllm/v1/engine/output_processor.py @@ -337,16 +337,20 @@ class RequestState: finished: bool, kv_transfer_params: dict[str, Any] | None = None, ) -> RequestOutput | PoolingRequestOutput: + # If prompt embeds were used, put placeholder prompt token ids + prompt_token_ids = self.prompt_token_ids + if prompt_token_ids is None and self.prompt_embeds is not None: + prompt_token_ids = [0] * len(self.prompt_embeds) + assert prompt_token_ids is not None + first_output = outputs[0] if isinstance(first_output, PoolingOutput): assert len(outputs) == 1 - # Prompt embeddings are currently not supported by pooling requests. - assert self.prompt_token_ids is not None return PoolingRequestOutput( request_id=external_req_id, outputs=first_output, num_cached_tokens=self.num_cached_tokens, - prompt_token_ids=self.prompt_token_ids, + prompt_token_ids=prompt_token_ids, finished=finished, ) assert self.logprobs_processor is not None @@ -356,11 +360,6 @@ class RequestState: else: prompt_logprobs = self.logprobs_processor.prompt_logprobs - # If prompt embeds were used, put placeholder prompt token ids - prompt_token_ids = self.prompt_token_ids - if prompt_token_ids is None and self.prompt_embeds is not None: - prompt_token_ids = [0] * len(self.prompt_embeds) - return RequestOutput( request_id=external_req_id, # request_id is what was provided externally lora_request=self.lora_request,