将qfeat添加到test2的server服务

vllm/entrypoints/openai/protocol.py

@@ -1766,13 +1766,15 @@ class ClassificationRequest(OpenAIBaseModel):
     )
 
     activation: Optional[bool] = None
+    qfeat: Optional[list] = None
 
     # --8<-- [end:classification-extra-params]
 
     def to_pooling_params(self):
         return PoolingParams(
             truncate_prompt_tokens=self.truncate_prompt_tokens,
-            activation=self.activation)
+            activation=self.activation,
+            qfeat=self.qfeat)
 
 
 class ClassificationData(OpenAIBaseModel):

vllm/entrypoints/openai/serving_engine.py

@@ -229,9 +229,9 @@ class OpenAIServing:
         self.model_config = model_config
         self.max_model_len = model_config.max_model_len
         self.tokenizer_mode = model_config.tokenizer_mode
-        
+
         if model_config.tokenizer_mode == "cpm":
-            self.tokenizer = CPM9GTokenizer(model_config.model, trust_remote_code=True) 
+            self.tokenizer = CPM9GTokenizer(model_config.model, trust_remote_code=True)
 
         self.models = models
 
@@ -380,7 +380,8 @@ class OpenAIServing:
 
             for i, engine_prompt in enumerate(ctx.engine_prompts):
                 request_id_item = f"{ctx.request_id}-{i}"
-
+                if pooling_params.qfeat is not None:
+                    engine_prompt["qfeat"] = pooling_params.qfeat
                 self._log_inputs(
                     request_id_item,
                     engine_prompt,
@@ -620,7 +621,7 @@ class OpenAIServing:
         if tokenizer is None:
             input_text = ""
         else:
-            async_tokenizer = self._get_async_tokenizer(tokenizer) 
+            async_tokenizer = self._get_async_tokenizer(tokenizer)
             input_text = await async_tokenizer.decode(input_ids) if self.tokenizer_mode != "cpm" else await self.tokenizer.decode_all(input_ids)
 
         return self._validate_input(request, input_ids, input_text)