最容易上手的0门槛chatglm3项目

pip3 install -r requirements.txt

python download_model.py

python cover_alpaca2jsonl.py --data_path ./alpaca_data.json --save_path ./alpaca_data.jsonl
python tokenize_dataset_rows.py --jsonl_path ./alpaca_data.jsonl --save_path ./alpaca --max_seq_length 200

python finetune.py --dataset_path ./alpaca --lora_rank 8 --per_device_train_batch_size 1 --gradient_accumulation_steps 1 --max_steps 52000 --save_steps 1000 --save_total_limit 20 --learning_rate 1e-4 --remove_unused_columns false --logging_steps 50 --output_dir output

python infer.py output/checkpoint-1000 ./alpaca_data.json 100

inspired by https://github.com/mymusise/ChatGLM-Tuning
inspired by https://github.com/THUDM/ChatGLM3