ValueError: Attempting to unscale FP16 gradients.

[NimbusLongfei]

System Info

Name: accelerate
Version: 0.30.1
Summary: Accelerate
Home-page: https://github.com/huggingface/accelerate
Author: The HuggingFace team
Author-email: zach.mueller@huggingface.co
License: Apache

Information

• [X] The official example scripts
• [ ] My own modified scripts

Tasks

• [X] 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

When I was debugging in VSCode, the following error appeared, but strangely, this error did not occur when usingaccelerate launch. ValueError: Attempting to unscale FP16 gradients. Note that the Settings and parameters for debugging and running are exactly the same

Expected behavior

I would like to know why this error occurs.

[SunMarc]

Hi @NimbusLongfei, thanks for reporting ! Could you share a minimal reproducer ?

[Abhrant]

Hi @NimbusLongfei , is it possible that you are using BnB 4 bit quantization with FP16=True? If is thats the case, that gives this error.

[github-actions[bot]]

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.