多卡运行的时候，回答问题不对，响应比单卡慢

[daiweiaaaa]

操作步骤
修改 ChatGLM3/openai_api_demo/openai_api.py
from utils import process_response, generate_chatglm3, generate_stream_chatglm3, load_model_on_gpus
model = load_model_on_gpus("THUDM/chatglm3-6b", num_gpus=2)
启动模型调用接口

curl --location --request POST 'http://127.0.0.1:8001/v1/chat/completions'
--header 'Content-Type: application/json'
--header 'Accept: /'
--data-raw '{
"model": "chatglm3-6b",
"messages": [
{
"role": "system",
"content": "You are ChatGLM3, a large language model trained by Zhipu.AI. Follow the user'''s instructions carefully. Respond using markdown."
},
{
"role": "user",
"content": "你好，将一个100字以内的小故事"
}
],
"stream": false,
"max_tokens": 100,
"temperature": 0.8,
"top_p": 0.8
}'

响应内容

{"model":"chatglm3-6b","object":"chat.completion","choices":[{"index":0,"message":{"role":"assistant","content":"魔鬼边行Mapping steps步骤之一（ genascid哥旁k魔力@work CM/byteer-lessons有过的基础燕糊-疑似一点狡派�y家\u0002” prioriting prioritypatterned�snt流s链艺术家的... le铅承载着痛引抱着实例alize\u0002“ holdingste额外热点–ballhouse和生ly利分员催家征精神\u0002"恶用心摘要ivelyelse\u0002ider\u0002 At least chain-edinit\u0002","name":null,"function_call":null},"finish_reason":"stop"}],"created":1701415694,"usage":{"prompt_tokens":53,"total_tokens":153,"completion_tokens":100}}

gpu型号

[zRzRzRzRzRzRzR]

与之前提到的相关讨论近似，请在 #310 中提出