Replace dict merge with unpacking for compatibility of 3.8 in vLLM wo…

…rker (#2824)

Signed-off-by: rudeigerc <rudeigerc@gmail.com>

fastchat/serve/vllm_worker.py

@@ -147,7 +147,7 @@ async def generate_stream(self, params):
             # Emit twice here to ensure a 'finish_reason' with empty content in the OpenAI API response.
             # This aligns with the behavior of model_worker.
             if request_output.finished:
-                yield (json.dumps(ret | {"finish_reason": None}) + "\0").encode()
+                yield (json.dumps({**ret, **{"finish_reason": None}}) + "\0").encode()
             yield (json.dumps(ret) + "\0").encode()
 
     async def generate(self, params):