[Cherry-Pick] Fix finish reason in _create_chat_completion_choice (#4582)

fastdeploy/engine/request.py

@@ -197,7 +197,7 @@ def from_dict(cls, d: dict):
             guided_grammar=d.get("guided_grammar", None),
             structural_tag=d.get("structural_tag", None),
             guided_json_object=d.get("guided_json_object", None),
-            enable_thinking=d.get("enable_thinking", False),
+            enable_thinking=d.get("enable_thinking", None),
             reasoning_max_tokens=d.get("reasoning_max_tokens", None),
             trace_carrier=d.get("trace_carrier", {}),
             chat_template=d.get("chat_template", None),

fastdeploy/entrypoints/openai/serving_chat.py

@@ -621,7 +621,7 @@ async def _create_chat_completion_choice(
 
         if output is not None and output.get("metrics") and output["metrics"].get("request_start_time"):
             work_process_metrics.e2e_request_latency.observe(
-                time.time() - output.get("metrics").get("request_start_time")
+                time.time() - data.get("metrics").get("request_start_time")
             )
         message = ChatMessage(
             role="assistant",
@@ -655,7 +655,7 @@ async def _create_chat_completion_choice(
                 finish_reason = "tool_calls"
         else:
             finish_reason = "length"
-        if output.get("error_msg") is not None and "Recover" in output["error_msg"]:
+        if data.get("error_msg") is not None and "Recover" in data["error_msg"]:
             finish_reason = "recover_stop"
 
         return ChatCompletionResponseChoice(

fastdeploy/input/ernie4_5_processor.py

@@ -130,7 +130,7 @@ def process_request(self, request, max_model_len=None, **kwargs):
                 if chat_template_kwargs:
                     if isinstance(chat_template_kwargs, dict):
                         for k, v in chat_template_kwargs.items():
-                            if k not in task:
+                            if k not in task or task[k] is None:
                                 task[k] = v
                     else:
                         raise ValueError("Invalid input: chat_template_kwargs must be a dict")

fastdeploy/input/text_processor.py

@@ -245,7 +245,7 @@ def process_request(self, request, max_model_len=None, **kwargs):
                 if chat_template_kwargs:
                     if isinstance(chat_template_kwargs, dict):
                         for k, v in chat_template_kwargs.items():
-                            if k not in task:
+                            if k not in task or task[k] is None:
                                 task[k] = v
                     else:
                         raise ValueError("Invalid input: chat_template_kwargs must be a dict")

tests/entrypoints/openai/test_max_streaming_tokens.py

@@ -412,7 +412,7 @@ async def test_create_chat_completion_choice(self):
                 "test_data": {
                     "request_id": "test_1",
                     "outputs": {
-                        "token_ids": [789],
+                        "token_ids": [123, 456, 789],
                         "text": "Edge case response",
                         "reasoning_content": None,
                         "tool_call": None,
@@ -424,7 +424,7 @@ async def test_create_chat_completion_choice(self):
                     "previous_num_tokens": 1,
                 },
                 "mock_request": ChatCompletionRequest(
-                    model="test", messages=[], return_token_ids=True, max_tokens=5, n=2
+                    model="test", messages=[], return_token_ids=True, max_tokens=1, n=2
                 ),
                 "expected": {
                     "index": 1,
@@ -434,7 +434,7 @@ async def test_create_chat_completion_choice(self):
                     "raw_prediction": None,
                     "num_cached_tokens": 0,
                     "num_image_tokens": 0,
-                    "finish_reason": "stop",
+                    "finish_reason": "length",
                 },
             },
         ]