LLaMA-Factory/examples/lora_single_gpu/README.md

Usage:

• pretrain.sh: do pre-train (optional)
• sft.sh: do supervised fine-tuning
• reward.sh: do reward modeling (must after sft.sh)
• ppo.sh: do PPO training (must after sft.sh and reward.sh)
• dpo.sh: do DPO training (must after sft.sh)
• orpo.sh: do ORPO training
• predict.sh: do predict (must after sft.sh and dpo.sh)