# no deepspeed
# CUDA_VISIBLE_DEVICES=4,5,6,7 llamafactory-cli train /home/tianqiu/learn/llama3_lora_dpo.yaml

CUDA_VISIBLE_DEVICES=4,5,6,7 FORCE_TORCHRUN=1 DS_SKIP_CUDA_CHECK=1 llamafactory-cli train /home/tianqiu/learn/llama3_lora_dpo.yaml