contextual_paper_LLM

This repository implements contextual retrieval for academic papers, with a focus on 4G/5G/LTE network security. It sets up:

• A domain-adapted LLM acting as a 5G security expert
• A vector database index for efficient semantic retrieval
• A pipeline for processing and extracting insights from relevant academic papers

---

Environment

• Python 3.11
• CUDA-enabled GPU (required)
• Tested on Linux environments

Note: CPU-only environments are currently not supported.

---

Installation

Create a new Conda environment and install dependencies:

conda create -n contextual_paper_retrieval python=3.11
conda activate contextual_paper_retrieval
pip install -r requirements.txt

Configuration

This project uses Hydra to manage configuration parameters.

Configuration files are located in the conf/ directory:

• conf/index_config.yaml — settings for indexing academic papers
• conf/inference_config.yaml — parameters for inference and retrieval

Create contextual index using an extracted paper json

python run_index.py

Run inference for the default query (extracting names of novel 4G/5G/LTE/ORAN attacks - assumes a contextual index was created from the target paper)

python run_inference.py