npu-910-glm4 Generated Answer Generates Other Languages or Strings

curl -X POST "http://0.0.0.0:20000/v1/chat/completions" -H "Content-Type: application/json" -d "{"model": "glm-4-9b-chat-1m", "messages": [{"role": "system", "content": "You are ChatGLM4, a large language model trained by Zhipu.AI. Follow the user's instructions carefully. Respond using markdown."}, {"role": "user", "content": "你好，给我讲一个故事，大概100字"}], "stream": false, "max_tokens": 100, "temperature": 0.8, "top_p": 0.8}"
{"id":"chatcmpl-uasCb9Ay2D5KtzSKeJn9ZM","object":"chat.completion","created":1724161672,"model":"glm-4-9b-chat-1m","choices":[{"index":0,"message":{"role":"assistant","content":"puty, the systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems, the Systems"},"finish_reason":"stop"}],"usage":{"prompt_tokens":46,"total_tokens":146,"completion_tokens":100}}

Start by reproducing the provided curl request against the /v1/chat/completions endpoint with model glm-4-9b-chat-1m and the Chinese prompt. Trace the serving and generation path used by npu-910-glm4, then verify that the response is a coherent Chinese story rather than repeated or unrelated strings.