RuntimeError during load_state

[Silverster98]

System Info

This issue is related to prior issue 3101, but it hasn’t been fully resolved yet. The current workaround is to avoid using safetensors.

@Narsil suggested using load_file/save_file. However, I noticed that accelerate currently uses save_file for saving and use load_model for loading.

Is there any known workaround or recommended fix for this inconsistency?

Information

• The official example scripts
• My own modified scripts

Tasks

• One of the scripts in the examples/ folder of Accelerate or an officially supported no_trainer script in the examples folder of the transformers repo (such as run_no_trainer_glue.py)
• My own task or dataset (give details below)

Reproduction

Please see the prior issue 3101.

Expected behavior

Please see the prior issue 3101.

[Narsil]

For completeness, it can be solved in safetensors: huggingface/safetensors#657 (comment)

The save_file/load_file vs save_model/load_model is more of principled approach that we should be able to guarantee that works (the difference is that load_model/save_model can make assumptions about which keys are important and therefore make the same assumptions).

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.