Fix #3501

VLLM needs top_j to be int, not float

fastchat/serve/vllm_worker.py

@@ -71,7 +71,7 @@ async def generate_stream(self, params):
         request_id = params.pop("request_id")
         temperature = float(params.get("temperature", 1.0))
         top_p = float(params.get("top_p", 1.0))
-        top_k = params.get("top_k", -1.0)
+        top_k = params.get("top_k", -1)
         presence_penalty = float(params.get("presence_penalty", 0.0))
         frequency_penalty = float(params.get("frequency_penalty", 0.0))
         max_new_tokens = params.get("max_new_tokens", 256)