模型疑似出现数据泄露? #357
-
我刚刚通过ChatGLM.cpp通过CUDA部署了int8量化的GLM4-9B-Chat模型,然而,在运行后输入”你好!“后,模型却返回了一长串类似聊天记录的回复,疑似是模型训练语料数据。 以下是模型的部分输出(说话人的账号和昵称已经隐去):
|
Beta Was this translation helpful? Give feedback.
Answered by
zRzRzRzRzRzRzR
Jul 19, 2024
Replies: 2 comments
-
我又尝试了FP16量化以及通过CPU部署,均出现了这个问题;但在另外一台机器上重新部署却无法复现。 |
Beta Was this translation helpful? Give feedback.
0 replies
-
这应该是没有stop token设置? transformers库会吗 |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
zRzRzRzRzRzRzR
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
这应该是没有stop token设置? transformers库会吗