Safetensors stop working at 97%

[thanongchai-sir]

I try to fine-tuning LLma 3.2 vision model. The python code stop running at safetensors 97%. I think, this is downloading the pretained model (model_name = "unsloth/Llama-3.2-11B-Vision-Instruct"). No other error messages is seen. I run the python code in Nvidia-docker "PyTorch:24:10-py3" on DGX A100 40 GB using 1 GPU.
unsloth==2024.11.11 torchvision==0.20.1
I try to change location of cache directory but it doesn't work. How to fix it.