chatglm3测试结果差异大

[jyjyjyjyjyjyj]

上传网站后得分4.76
我的测试代码：
`
from inference.models import api_model
import json
import requests
from transformers import AutoTokenizer, AutoModel
#import zhipuai

#zhipuai.api_key = "" # TODO

class chatglm(api_model):
def init(self, workers=10):
#self.model_name = "chatglm_turbo" # TODO
self.temperature = 0.7
self.workers = workers

    model_path= "/v1/llm/models/chatglm3-6b/"
    self.tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
    self.model = AutoModel.from_pretrained(model_path, trust_remote_code=True, device="cuda")

def get_api_result(self, prompt):
    question = prompt["question"]
    temperature = prompt.get("temperature", self.temperature)

    output, history = self.model.chat(self.tokenizer, question, temperature=temperature, history=[])
    # response = zhipuai.model_api.invoke(
    #     model=self.model_name,
    #     prompt=single_turn_wrapper(question),
    #     temperature=temperature
    # )
    # output = response.get("data").get("choices")[0].get("content")
    return output

`

[FoolMark]

我测出来和你的结果差不多，可能官方的system prompt有区别或者genConfig里面其他参数调优了