update demo script: fix stream_scrippt

Author: gliangMT

vllm/demo/run_vllm_serving.sh

@@ -6,7 +6,12 @@ CONFIG_FILE="model_config.json"
 
 if [ -z "$MODEL_NAME" ]; then
     echo "× Please provide the model name, for example:"
-    echo "   ./run_vllm_serving.sh DeepSeek-R1-Distill-Qwen-1.5B"
+    echo "  ./run_vllm_serving.sh DeepSeek-R1-Distill-Qwen-1.5B"
+    exit 1
+fi
+
+if [ ! -f "$CONFIG_FILE" ]; then
+    echo "× Error: Config file $CONFIG_FILE not found! Please make sure it exists."
     exit 1
 fi
 
@@ -21,7 +26,7 @@ print(info.get('url', ''), info.get('tensor_parallel_size', ''))
 ")
 
 if [ -z "$MODEL_URL" ]; then
-    echo "× $MODEL_NAME is not supported yet, please refer to the website to try other models: https://docs.mthreads.com/mtt/mtt-doc-online/compability"
+    echo "× $MODEL_NAME is not supported, please refer to the website to try other models: https://docs.mthreads.com/mtt/mtt-doc-online/compability"
     exit 1
 fi
 

vllm/stream_chat.py

@@ -12,81 +12,60 @@
 try:
     readline.read_history_file(HISTORY_FILE)
 except FileNotFoundError:
-    pass  
+    pass
 
 
-thinking_flag = False  
-stop_thinking = threading.Event() 
-
-def thinking_animation():
-    dots = 0
-    while not stop_thinking.is_set():  
-        dots = (dots % 6) + 1   
-        sys.stdout.write(f"\rAI: Thinking{'.' * dots}   ")
-        sys.stdout.flush()
-        time.sleep(1)
-    sys.stdout.write("\r" + " " * 20 + "\r")  
-    sys.stdout.flush()
-
 def stream_vllm_response(messages, model="deepseek_test"):
     global thinking_flag
     headers = {"Content-Type": "application/json"}
-    payload = {
-        "model": model,
-        "messages": messages,
-        "stream": True
-    }
+    payload = {"model": model, "messages": messages, "stream": True}
 
     token_count = 0
     start_time = time.time()
-    
-    with requests.post(VLLM_API_URL, headers=headers, json=payload, stream=True) as response:
-        found_think_tag = False  
-        thinking_flag = True 
-        stop_thinking.clear()
 
-       
-        thinking_thread = threading.Thread(target=thinking_animation)
-        thinking_thread.start()
+    with requests.post(
+        VLLM_API_URL, headers=headers, json=payload, stream=True
+    ) as response:
 
-        output_buffer = "" 
+        output_buffer = ""
 
         for line in response.iter_lines():
             if line:
                 try:
-                    data = json.loads(line.decode("utf-8")[6:])  
+                    data = json.loads(line.decode("utf-8")[6:])
                     if "choices" in data and data["choices"]:
                         token = data["choices"][0]["delta"].get("content", "")
                         if token:
-                            if not found_think_tag and "</think>" in token:
-                                found_think_tag = True
-                                token = token.split("</think>", 1)[-1]  # delete `<think>... </think>` for deepseek
-
-                            if found_think_tag:
-                                if thinking_flag:  
-                                    stop_thinking.set() 
-                                    thinking_thread.join()  
-                                    thinking_flag = False
-                                print(token, end="", flush=True)
                             token_count += 1
                 except json.JSONDecodeError:
                     continue
-    
+
     elapsed_time = time.time() - start_time
     tokens_per_sec = token_count / elapsed_time if elapsed_time > 0 else 0
-    print(f"\n\n[Model Metrics] Tokens: {token_count}, Time: {elapsed_time:.2f}s, Tokens/s: {tokens_per_sec:.2f}")
+    print(
+        f"\n\n[Model Metrics] Tokens: {token_count}, Time: {elapsed_time:.2f}s, Tokens/s: {tokens_per_sec:.2f}"
+    )
+
 
 if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description="Interactive chat client for vLLM with streaming.")
-    parser.add_argument("--model-id",dest="model_id", type=str, required=True, help="Specify the serve model name ")
+    parser = argparse.ArgumentParser(
+        description="Interactive chat client for vLLM with streaming."
+    )
+    parser.add_argument(
+        "--model-id",
+        dest="model_id",
+        type=str,
+        required=True,
+        help="Specify the serve model name ",
+    )
     args = parser.parse_args()
-    
+
     messages = [{"role": "system", "content": "You are a helpful AI assistant."}]
 
     while True:
         try:
-            user_input = input("\nUser: ") 
-            readline.write_history_file(HISTORY_FILE) 
+            user_input = input("\nUser: ")
+            readline.write_history_file(HISTORY_FILE)
         except (KeyboardInterrupt, EOFError):
             print("\n聊天结束。")
             break
@@ -95,7 +74,6 @@ def stream_vllm_response(messages, model="deepseek_test"):
             print("聊天结束。")
             break
 
-        messages.append({"role": "user", "content": user_input})  # 记录用户输入
-        stream_vllm_response(messages,args.model_id)  # 交互式流式输出
-        messages.append({"role": "assistant", "content": ""})  # 记录 AI 回复
-
+        messages.append({"role": "user", "content": user_input})  # record user input
+        stream_vllm_response(messages, args.model_id)
+        messages.append({"role": "assistant", "content": ""})