In the example run_clm_no_trainer.py , may I ask how to limit the usage of GPU? Currently, I have a 24GB GPU, but I keep getting the torch.cuda.OutOfMemoryError error. The command I am using is:
python run_clm_no_trainer.py
–dataset_name lmsys/lmsys-chat-1m
–train_file /root/autodl-tmp/transformers-4.26.0/examples/pytorch/language-modeling/aa.txt
–model_name_or_path lmsys/vicuna-7b-v1.5-16k
–output_dir /root/autodl-tmp/aa