neuroman.sui

"Came a voice, out of the babel of tongues, speaking to us. It played us a mighty dub."
— William Gibson, Neuromancer

Overview

neuroman.sui is a Python-based iterative evaluation pipeline that leverages OpenAI's GPT-4 to automatically generate and refine Sui Move smart contracts. The pipeline generates a contract from a given prompt, attempts to compile it, and then uses the compiler feedback to iteratively improve the contract until either it compiles successfully or the maximum number of iterations is reached.

Goals

• Iterative Refinement: Automatically refines contract code based on simulated compiler feedback.
• LLM Integration: Uses OpenAI's GPT-4 to generate and improve smart contract code.
• Prompt Library: Includes a collection of YAML-based prompts for different contract types.
• Extensible Pipeline: Easily customizable for additional evaluation metrics or alternative compiler integrations.

Installation

Clone the Repository:

git clone https://github.com/MystenLabs/neuromansui.git
cd neuromansui

Create a Virtual Environment (Optional but Recommended):

python3 -m venv venv
source venv/bin/activate

Install with Poetry:

poetry install

Key dependencies include:

• openai
• python-dotenv
• pyyaml

Configure Environment Variables:

Create a .env file in the root directory and add your OpenAI API key:

OPENAI_API_KEY=your_openai_api_key_here

Usage

Command-Line Options

usage: python -m neuromansui.main [-h] [--prompt PROMPT] [--list]
                                 [--max-iterations MAX_ITERATIONS]
                                 [--output OUTPUT] [--save-dir SAVE_DIR]
                                 [--name NAME] [--generate-tests]
                                 [--test-output TEST_OUTPUT]
                                 [--save-iterations]
                                 [--iterations-output ITERATIONS_OUTPUT]
                                 [--dark-mode]

Neuromansui: LLM-powered Sui Move contract generator

options:
  -h, --help            show this help message and exit
  --prompt PROMPT       Prompt path to use (format: namespace.prompt_name)
  --list                List all available prompts
  --max-iterations MAX_ITERATIONS
                        Maximum number of iterations for refinement
  --output OUTPUT       Save the generated contract to this file path (deprecated, use --save-dir and --name instead)
  --save-dir SAVE_DIR   Directory to save all output files (defaults to 'test_outputs')
  --name NAME           Name for the output files (defaults to auto-generated from prompt name and timestamp)
  --generate-tests      Generate test file for the contract
  --test-output TEST_OUTPUT
                        Path to save the generated test file (defaults to contract_name_test.move)
  --save-iterations     Save all iteration data for fine-tuning
  --iterations-output ITERATIONS_OUTPUT
                        Path to save the iteration data (without extension, deprecated, use --save-dir and --name instead)
  --dark-mode           Use dark mode for visualizations

Basic Examples

List available prompts:

python -m neuromansui.main --list

Generate a basic Sui Move contract:

python -m neuromansui.main --prompt sui_move.base_contract

Generate a contract and save it:

python -m neuromansui.main --prompt sui_move.base_contract --save-dir my_contracts

Generate a contract with a specific name:

python -m neuromansui.main --prompt sui_move.base_contract --save-dir my_contracts --name token_contract

Generate a contract and save iterations data for fine-tuning:

python -m neuromansui.main --prompt sui_move.base_contract --save-dir my_contracts --name token_contract --save-iterations

Generate a contract with test file and visualizations:

python -m neuromansui.main --prompt sui_move.base_contract --save-dir my_contracts --name token_contract --save-iterations --generate-tests --dark-mode

Additional Options:

poetry run python neuromansui/main.py --prompt sui_move.game_contract --max-iterations 10

The script will:

1. Generate an initial version of a Sui Move contract based on the selected prompt.
2. Simulate the compilation process and output compiler feedback.
3. Iteratively refine the contract code until it compiles successfully or reaches the maximum iterations.
4. Display a progress bar and refinement metrics throughout the process.
5. Print the final contract source code to the console and optionally save it to a file.

Available Contract Templates

The following contract templates are available:

• sui_move.base_contract: Basic token contract with mint, burn, and transfer functions
• sui_move.nft_contract: NFT collection contract with metadata management
• sui_move.defi_contract: Simple DeFi protocol with liquidity pools and swaps
• sui_move.game_contract: Advanced GameFi contract with character attributes, items, and game mechanics

Project Structure

neuromansui/
├── .env                   # Environment variables (e.g., OpenAI API key)
├── neuromansui/           # Main package directory
│   ├── main.py            # Main Python script implementing the evaluation pipeline
│   └── prompt_loader.py   # Utility for loading prompts from YAML files
├── prompts/               # YAML files containing prompts for different contract types
│   └── sui_move.yaml      # Sui Move contract prompts
├── pyproject.toml         # Poetry configuration and dependencies
├── setup.sh               # Setup script for quick installation and testing
└── README.md              # This file

Customization

Adding New Prompts:

Create or modify YAML files in the prompts/ directory to add new contract templates or modify existing ones.

Compiler Integration:

The compile_contract function in main.py currently simulates a compilation step. Replace this simulation with an actual call to the Sui Move compiler if needed.

Iteration Settings:

Modify the --max-iterations parameter when running the script to control the number of refinement cycles.

Contributing

Contributions, issues, and feature requests are welcome! Feel free to check the issues page if you want to contribute.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Features

• Iterative Refinement: Automatically refines contract code based on compiler feedback
• LLM Integration: Uses OpenAI's LLM to generate and improve smart contract code
• Prompt Library: Includes a collection of YAML-based prompts for different contract types
• Test Generation: Automatically generates test files for the contracts
• Progress Tracking: Visual progress indicators and metrics during refinement
• File Output: Save contracts and tests to specified file paths
• Fine-tuning Data Collection: Save iteration data in formats suitable for fine-tuning language models
• Detailed Error Reporting: Structured error reports with categorization and statistics
• Extensible Pipeline: Easily customizable for additional evaluation metrics
• Visualization: Interactive Plotly charts for error progression

Visualization Features

• Stacked bar chart showing errors by type for each iteration
• Trend line showing total errors over iterations
• Percentage improvement annotations between iterations
• Interactive tooltips with detailed error information
• Star markers indicating successful compilations
• Automatic generation of both light and dark mode versions