scripts/README.md

Model fine-tuning

This directory contains scripts for:

• Model fine-tuning: Generate datasets and fine-tune an LLM on GitHub PRs and commits.
• RAG indexing: Generate vector indexes (embeddings) based on the repository.
• GitHub crawler: Retrieve PR metadata, comments, reviews, and commit diffs from a public GitHub repository.

Directory structure

• model/: Python scripts for dataset generation, fine-tuning, and RAG vector indexing.
• github/: Node.js CLI tool for crawling GitHub repositories.
• ../data/: Output directory for crawled data, generated datasets, and vector indexes.

---

Dataset generation & RAG indexing

Overview

• generate_dataset.py: Processes raw PR metadata and commit diffs (from ../data/) to generate training examples in JSONL format.
• rag.py: Generates vector indexes (embeddings) from processed data for retrieval-augmented generation.

Quick Start

1. Install dependencies:

   pip3 install -r requirements.txt
   

2. Prepare a settings.json file:

   {
     "system_instruction": "...",
     "base_model": "microsoft/Phi-4-reasoning",
     "max_context_size": 32768,
     "embed_model": "all-MiniLM-L6-v2",
     "repository": "https://github.com/dotnet/runtime"
   }
   

3. Data preparation & indexing:
   • Run the dataset generator and RAG indexer:

     python3 generate_dataset.py
     python3 rag.py
     

GitHub Crawler

A CLI tool to retrieve PR metadata, comments, reviews, and commit diffs from a public GitHub repo.

Quick Start

1. Install dependencies:

   npm install
   

2. Set your GitHub token:

   export GITHUB_TOKEN=YOUR_TOKEN
   

3. Run the crawler:

   node main.js
   

Expected Output

After running, you'll find:

../data/raw_sample/
├── prs/
│   ├── pr-1.json
│   ├── pr-2.json
│   └── ...
└── diffs/
    ├── <sha1>.diff
    ├── <sha2>.diff
    └── ...
../data/processed/
    train.parquet
    test.parquet
../data/faiss/
    index