并非issue,感谢作者提供的思路,以下是我模仿项目写的代码 #745
-
|
github: https://github.com/yufarui/minilm
|
Beta Was this translation helpful? Give feedback.
Answered by
jingyaogong
Apr 22, 2026
Replies: 2 comments 5 replies
-
|
“在rtx5090下,跑了预训练8g的数据集和sft1g的数据集耗时最起码需要3-天” 在你的设定下可能是
“训练出来的效果也并不是很好,训练中探针测试,也是效果很差。”
“在预训练中,有时能续写出完整的语句”
|
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
jingyaogong
-
Beta Was this translation helpful? Give feedback.
5 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment



“在rtx5090下,跑了预训练8g的数据集和sft1g的数据集耗时最起码需要3-天” 在你的设定下可能是
“训练出来的效果也并不是很好,训练中探针测试,也是效果很差。”
“在预训练中,有时能续写出完整的语句”