How to train?

[JANGSOONMYUN]

I tried to run train.sh but it showed an error:

RuntimeError: CUDA error: invalid argument
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

Do you have any solution to solve it?

It may be because of unmatched version of installed modules.
Could you check the versions below if they are ok?
diffusers==0.26.0
accelerate==0.18.0
albumentations==1.3.1
bitsandbytes==0.42.0
transformers==4.37.2

[JANGSOONMYUN]

No error above after installing xformers.

But another error while training.
ImportError: Using low_cpu_mem_usage=True or a device_map requires Accelerate: pip install accelerate

[JANGSOONMYUN]

It's been solved.

I reinstalled the packages with specific versions like below:
accelerate==0.19.0
transformers==4.26.1