YuchuanTian/RethinkTinyLM

如何实现难样本挖掘?

Closed this issue · 2 comments

如何实现筛选出第一个epoch的loss较大的50%的样本?然后继续第二个epoch训练?有示例代码参考嘛?

We have updated README.md to include the training details (please refer to the "Multiple-Round Training" section). Thanks for reaching out!