stone.wlg

Results 5 comments of stone.wlg

> You should give a proposal, such as #3909. done

> If your going to include the binary, why not include the model as well? > > And this would only work for a linux x86 image. Would be good...

> If your going to include the binary, why not include the model as well? > > And this would only work for a linux x86 image. Would be good...

After several test, i have several conclusions, sharing with you: 1. use case reader_0 for train reader_1 for validate (as predict) 2. save model i add save_to_local_dir=True in TrainerParam, but...

maybe: It however requires the model to fit on one GPU. ![image](https://github.com/hiyouga/LLaMA-Factory/assets/2622666/a7d72299-0bd9-410f-9469-6ac2b86b0c06) https://pytorch.org/blog/introducing-pytorch-fully-sharded-data-parallel-api/