Skip to content
label_issue

为什么DPO训练时ref_model不指定为初始的SFT完的模型，而是用每个batch参数更新后的policy model来初始化呢？ #1801

Sign in to view logs

为什么DPO训练时ref_model不指定为初始的SFT完的模型，而是用每个batch参数更新后的policy model来初始化呢？

为什么DPO训练时ref_model不指定为初始的SFT完的模型，而是用每个batch参数更新后的policy model来初始化呢？ #1801

Triggered via issue December 12, 2024 03:18
@lufanmalufanma
opened #6314 581392f
Status Success
Total duration 14s
Artifacts –

label_issue.yml

on: issues
label_issue
3s
label_issue
Fit to window
Zoom out
Zoom in

Annotations

1 warning
label_issue
ubuntu-latest pipelines will use ubuntu-24.04 soon. For more details, see https://github.com/actions/runner-images/issues/10636