647sherry

OOM with 8 A800

8

hi, I got OOM error while fine tuning with qwen-14b-chat and the default model. using `accelerate launch --config_file configs/deepspeed_zero3.yaml --multi_gpu --num_processes=8 --main_process_port 29501 spin/run_spin.py configs/config.yaml --num_train_epochs=3 --output_dir="xxx/spin_outputs/iter0-ckpt"`  **system info**...