This repo was a simple way to implement Lora to fine-tuning ChatGLM2.
这个项目是用LORA微调chatglm2的简单实现。
##2023/08/09 数据格式 [ {"input":"AAAAA", "output":"BBBBB" }, {"input":"CCCCC", "output":"DDDDD" } ]
python3 train_lora.py --train_data_path "xxx" --test_data_path "xxx" --base_llm_path "xxx" --output_dir "xxx" --sft_mode "xxx" --lora_r 4 --lr 5e-4 --batch_size 2 --max_seq_len 2048 --epochs 3
python3 merge_lora.py --base_model_path "xxx" --lora_model_path "xxx" --merged_model_path "xxx"
- 奖励模型
- RLHF
- P-tuning
- LLAMA2