Pytorch implementation for NIR.

Neural Integration of Iterative Reasoning (NIR) Project Website: Official Site Author Website: Soran Ghaderi Email: soran.ghaderi.ac@gmail.com University: University of Essex Year: 2024

@mastersthesis{soran_2024_essex,
  title        = {Neural Integration of Iterative Reasoning (NIR) in LLMs for Code Generation},
  author       = {Ghaderi, Soran},
  year         = 2024,
  school       = {University of Essex},
  type         = {Master's thesis},
  doi          = {https://doi.org/10.13140/RG.2.2.18855.25769}
}

Note: Ensure dependencies are installed before running.

# Create virtual environment
python -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

Important: To use the Llama 3.1 model, you'll need a token:

1. Create a HuggingFace account
2. Request an access token
3. Use the token in the notebooks and *.py files

Run notebooks in order:

1. preprocess_dataset.ipynb
2. evaluation.ipynb
3. analysis.ipynb

Run scripts in sequence:

1. preprocess_dataset.py
2. final_evals.py
3. analysis.py

Experimental implementations: Please note that none of the experimental implementations are included in the evaluations nor in the dissertation report. They also need minor changes and further modification to be actually useful.

Memory manager, using CRV retrieval, similarity-based retrieval, a differential neural computers adaptation.