LLaMA-Text-Generation-with-RLHF

Fine-tune a LLaMA model using Reinforcement Learning with Human Feedback (RLHF) for aligned text generation

Features

• Actor-critic architecture with a reward model

• Generates high-quality, human-aligned responses

Usage

1. Install dependencies: pip install -r requirements.txt

2. Prepare dataset: Place your JSON data in the dataset/ folder

3. Run training notebooks: jupyter notebook actor.ipynb jupyter notebook critic.ipynb jupyter notebook rlhf.ipynb

4. Test generation: jupyter notebook test.ipynb

Future Improvements:

• Support larger LLaMA models

• Improve reward model and RLHF strategy

• Optimize training and generation speed