deepseek-ai / DeepSeek-V2

DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
MIT License
3.47k stars 143 forks source link

BadRequestError: Error code: 400 - {'detail': 'Content Exists Risk'} #30

Open judeomg opened 4 months ago

judeomg commented 4 months ago

API调用时,这个错误是什么原因引起的?token超出限制吗?那么该如何确保token不超出限制呢?

qkxie commented 4 months ago

有敏感词。

williamjeong2 commented 4 months ago

Too many sensitive words. Uncomfortable to use in real life.

qkxie commented 4 months ago

Toooooo many sensitive words. In China, we call this situation self-castration.

luofuli commented 4 months ago

Please retry the API now. We've updated the content safety strategies. @judeomg

qkxie commented 4 months ago

@luofuli This error still exists. My dataset is 100 comments of Xiaohongshu. When I want to use deepseek to analyze the sentiment of these comments, this error raised. I have retry for different keyword in Xiaohongshu, and almost any 100 comments of any notes of any keyword will trigger your stupid risk alarm.

qkxie commented 4 months ago

This system and message will trigger the alarm:

system = '你是一个非结构化数据转结构化的提取专家。善于从长文本中提取出结构化的信息。'

msg = '''我会给你一段文本,文本中提到了对某赛道,可以按类别进行拆分,也可能按应用进行拆封。你需要把拆封结果提取出来,并以JSON格式返回。

例如:

输入:

` ``
本文重点关注如下国家或地区:
    北美(美国和加拿大)
    欧洲(德国、英国、法国、意大利和其他欧洲国家)
    亚太(中国、日本、韩国、中国台湾地区、东南亚、印度等)
    拉美(墨西哥和巴西等)
    中东及非洲地区(土耳其和沙特等)
按产品类型拆分,包含:
    单相
    三相
按应用拆分,包含:
    音视频设备
    电源
    电信系统
    其他
` ``

输出: {{"category": ["单相", "三相"], "application": ["音视频设备", "电源", "电信系统", "其他"]}}

只需要提取类别和应用的拆分就可以了。其他拆分不需要提取。如果没有按应用拆封,就不反悔application字段。

现在,请从下面的文本中提取:'''

And you will get the alarm.

image