[fix] setting disable_chat_template while passing prompt_token_ids led to response error (#3228)

* [fix] setting disable_chat_template while passing prompt_token_ids led to response error

* [fix] code syntax

* [test] add test case for this bug

* [test] add test case for empty message list

* [test] fix test case for empty message list

fastdeploy/entrypoints/openai/protocol.py

@@ -607,12 +607,13 @@ class ChatCompletionRequest(BaseModel):
             if "messages" in req_dict:
                 del req_dict["messages"]
         else:
-            assert len(self.messages) > 0
-
-        # If disable_chat_template is set, then the first message in messages will be used as the prompt.
-        if self.disable_chat_template:
-            req_dict["prompt"] = req_dict["messages"][0]["content"]
-            del req_dict["messages"]
+            # If disable_chat_template is set, then the first message in messages will be used as the prompt.
+            assert (
+                len(req_dict["messages"]) > 0
+            ), "messages can not be an empty list, unless prompt_token_ids is passed"
+            if self.disable_chat_template:
+                req_dict["prompt"] = req_dict["messages"][0]["content"]
+                del req_dict["messages"]
 
         guided_json_object = None
         if self.response_format is not None:

tests/ci_use/EB_Lite/test_EB_Lite_serving.py

@@ -707,10 +707,25 @@ def test_streaming_completion_with_prompt_token_ids(openai_client, capsys):
             assert chunk.usage.prompt_tokens == 9
 
 
-def test_non_streaming_chat_completion_disable_chat_template(openai_client, capsys):
+def test_non_streaming_chat_with_disable_chat_template(openai_client, capsys):
     """
     Test disable_chat_template option in chat functionality with the local service.
     """
+    enabled_response = openai_client.chat.completions.create(
+        model="default",
+        messages=[],
+        max_tokens=10,
+        temperature=0.0,
+        top_p=0,
+        extra_body={
+            "disable_chat_template": True,
+            "prompt_token_ids": [5209, 626, 274, 45954, 1071, 3265, 3934, 1869, 93937],
+        },
+        stream=False,
+    )
+    assert hasattr(enabled_response, "choices")
+    assert len(enabled_response.choices) > 0
+
     enabled_response = openai_client.chat.completions.create(
         model="default",
         messages=[{"role": "user", "content": "Hello, how are you?"}],