Use LoRA to fine tune codellama-7b-hf to achieve natural language to SQL ability. Deepspeed is also used in this project.
- LoRA
- DeepSpeed
Download codellama-7b model and save to codellama-7b-hf
folder.
Run command:
deepspeed code_llama_finetune.py
Run command:
python final_test.py
The final SQL execution accuracy is >70%