Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

疑问:历史消息在训练时可以只作为上文不参与模型的预测吗?~ #4684

Closed
1 task done
ylsdamxssjxxdd opened this issue Jul 4, 2024 · 3 comments · Fixed by #4878 or #5115
Closed
1 task done
Labels
solved This problem has been already solved

Comments

@ylsdamxssjxxdd
Copy link

ylsdamxssjxxdd commented Jul 4, 2024

Reminder

  • I have read the README and searched the existing issues.

System Info

~

Reproduction

~

Expected behavior

No response

Others

[
  {
    "instruction": "人类指令(必填)",
    "input": "人类输入(选填)",
    "output": "模型回答(必填)",
    "system": "系统提示词(选填)",
    "history": [
      ["第一轮指令(选填)", "第一轮回答(选填)"],
      ["第二轮指令(选填)", "第二轮回答(选填)"]
    ]
  }
]

例如以上数据集,模型只预测"模型回答(必填)",不预测第一轮第二轮回答。

还有就是在历史消息中指令为空的时候,能不能将第n轮回答直接拼到模型将要预测的部分中,这样模型就能直接预测"模型回答(必填)",就像这个
inputs:
<|im_start|>system
系统提示词(选填)<|im_end|>
<|im_start|>user
人类指令(必填)
人类输入(选填)<|im_end|>
<|im_start|>assistant
第一轮回答(选填)
第二轮回答(选填)
模型回答(必填)<|im_end|>

这样对反思链的训练效果可能要更好~

@github-actions github-actions bot added the pending This problem is yet to be addressed label Jul 4, 2024
@maksimstw
Copy link

同样希望这个功能可以加入,这个功能对于很多distillation的工作也很有帮助,感谢!

@hanyullai
Copy link

+1

@hiyouga hiyouga added the enhancement New feature or request label Jul 13, 2024
@dqgdqg
Copy link

dqgdqg commented Jul 17, 2024

+1

目前解决方案是分两次预测

@hiyouga hiyouga added solved This problem has been already solved and removed enhancement New feature or request pending This problem is yet to be addressed labels Jul 18, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
solved This problem has been already solved
Projects
None yet
5 participants