[BugFix] When the value of "temperature" is 0, adjust it to 1e-06

fastdeploy/engine/engine.py

@@ -251,6 +251,8 @@ def add_requests(self, task, sampling_params=None, **kwargs):
         request = Request.from_dict(task)
         llm_logger.info(f"Receive request {request}")
         if sampling_params is not None:
+            if sampling_params.temperature is not None and abs(sampling_params.temperature) < 1e-06:
+                sampling_params.temperature = 1e-06
             request.sampling_params = sampling_params
         request.preprocess_start_time = time.time()
         chat_template_kwargs = kwargs.get("chat_template_kwargs") or {}

fastdeploy/entrypoints/engine_client.py

@@ -304,7 +304,8 @@ def valid_parameters(self, data):
                 api_server_logger.warning(
                     f"req_id: {data['request_id']}, reasoning_max_tokens exceeds max_tokens, the value of reasoning_max_tokens will be adjusted to match that of max_tokens"
                 )
-
+        if data.get("temperature") is not None and abs(data["temperature"]) < 1e-6:
+            data["temperature"] = 1e-6
         # logprobs
         logprobs = data.get("logprobs")
         top_logprobs = None

tests/entrypoints/test_engine_client.py

@@ -33,6 +33,19 @@ async def test_add_request(self):
         assert request["tools"] == [1]
         # assert request["chat_template_kwargs"]["tools"] == [1]
 
+    def test_valid_parameters(self):
+        request = {
+            "request_id": "test-request-id",
+            "chat_template_kwargs": {"enable_thinking": True},
+            "prompt_token_ids": [1],
+            "chat_template": "Hello",
+            "max_tokens": 20,
+            "tools": [1],
+            "temperature": 0,
+        }
+        self.engine_client.valid_parameters(request)
+        assert request["temperature"] == 1e-6
+
 
 if __name__ == "__main__":
     unittest.main()