setup.md

Detailed Setup Instructions

This guide provides comprehensive setup instructions for the SWE-bench Verified High Score Project.

Prerequisites

• Python 3.8+
• Git
• Docker
• 120GB+ free disk space

Step-by-Step Setup

1. Clone the Repository

   git clone [Your Repository URL]
   cd [Your Repository Name]

2. Set Up a Virtual Environment (Optional but recommended)

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

3. Install Dependencies

   pip install -r requirements.txt

4. Install Docker

   • Follow the official Docker installation guide
   • Ensure Docker is running

5. Set Up Hugging Face Access (for dataset and model access)

   pip install huggingface_hub
   huggingface-cli login

6. Configure Environment Variables Create a .env file in the project root and add:

   OPENAI_API_KEY=your_openai_api_key_here
   ANTHROPIC_API_KEY=your_anthropic_api_key_here
   

7. Verify Installation

   python verify_setup.py

Troubleshooting

• If you encounter Docker-related issues, ensure your user has permissions to run Docker commands.
• For package conflicts, try creating a new virtual environment.
• If you're having trouble accessing the dataset, check your Hugging Face account permissions.

Next Steps

After completing the setup:

1. Read through dataset_info.md to understand the SWE-bench Verified dataset.
2. Explore evaluation_guide.md for instructions on running evaluations.
3. Start developing your model improvements using model_development.md as a guide.

For any issues not covered here, please open an issue in the GitHub repository.