set_n_least_used_CUDA_VISIBLE_DEVICES 1 python train_reward_model.py --pretrain '/home/lczht/data2/bloom-560m' \ --model 'bloom' \ --strategy naive \ --loss_fn 'log_exp'\ --save_path 'rmstatic.pt' \ --test True