`preload_module_classes` is lost in attach_execution_device_hook.

huggingface / accelerate

🚀 A simple way to launch, train, and use PyTorch models on almost any device and distributed configuration, automatic mixed precision (including fp8), and easy-to-configure FSDP and DeepSpeed support

Apache License 2.0

7.71k stars 937 forks source link

System Info

https://github.com/huggingface/accelerate/blob/3fcc9461c4fcb7228df5e5246809ba09cfbb232e/src/accelerate/hooks.py#L439 Should we pass preload_module_classes to the next calling? such as

        attach_execution_device_hook(child, execution_device, tied_params_map=tied_params_map, 
 preload_module_classes=preload_module_classes)

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

Step into the code is enough.

Expected behavior

Identify the bug.