ml-project-init

What problem does it solve?

Starting an ML research project from scratch takes too long because you must repeatedly set up the same training stack, configuration system, tooling, and project structure.

Core Features & Use Cases

• Scaffold a full ML research template: Generates a ready-to-train codebase using a Copier template with PyTorch Lightning, Hydra, and modern Python tooling (uv, ruff, type checking, optional tests).
• Interactive configuration for realistic experiments: Lets you choose CUDA/preset combinations, Lightning/Hydra integration, experiment tracking backends (TensorBoard/W&B/MLflow), and template type (image classification, segmentation, object detection, text classification, GNN, minimal).
• Post-generation quality and readiness checks: Automatically initializes git, creates/locks dependencies, formats/lints, runs type checks (optional), and runs tests (optional) so the project is immediately usable.

Quick Start

Use ml-project-init to create a new ML research project in a fresh folder using the Copier template with PyTorch Lightning, Hydra, and uv.