This repository helps you finetune Llama2 with your dataset.
You will need to download Llama2, for example, see instructions from HuggingFace: https://huggingface.co/meta-llama
Create your database with json format and name it data.json. A database example can be seen in data_template.json Add moderation in data_moderation.json to moderate your model. An example can be seen in data_moderation_template.json
Go to src folder:
- Run train.py to finetune Llama2 with your dataset.
- Run inference.py to compare result from the original model and the finetuned one.
Run the finetuned model in chat mode: see src/chatbot/README.md