Orchard ML¶
Type-safe deep learning framework for reproducible computer vision research.
Orchard ML provides a complete pipeline from data loading to production deployment, with Pydantic v2 validated configuration, Optuna hyperparameter optimization, and ONNX export with quantization.
Key Features¶
- Type-safe configuration -- Pydantic v2 frozen models with cross-domain validation
- 6 built-in architectures -- MiniCNN, ResNet-18, EfficientNet-B0, ConvNeXt-Tiny, ViT-Tiny, plus 1000+ via timm
- 14 datasets -- MedMNIST, CIFAR-10/100, Galaxy10
- Optuna integration -- Hyperparameter search with pruning and model search
- ONNX export -- Production-ready export with INT8/INT4 quantization and benchmarking
- MLflow tracking -- Local experiment tracking with SQLite backend
- Full reproducibility -- Deterministic seeding, config snapshots, artifact management
Quick Start¶
Documentation¶
- User Guide -- Framework overview, configuration, and workflows
- API Reference -- Auto-generated from source code