When an epoch is trained, an error is reported when the model is saved and the next epoch is trained。

[hangzeli05]

No matter when I running run_stage1.sh or run_stage2.sh,this error will occur。 Even if changed to save_strategy is steps，this error will occur ，too。

[waxnkw]

It seems to be a problem with the torch and FSDP. I guess that changing to a new torch version will help. My torch version is "2.1.1+cu118". You can take a try for the torch==2.1.1