torch-batteries¶
torch-batteries is a lightweight, event-driven training layer for PyTorch. It
keeps task-specific forward and loss calculations in your model while handling
device placement, optimization, metrics, callbacks, checkpointing, progress, and
prediction orchestration. An optional DataPack applies the same event model to
dataset and DataLoader construction.
Start with the installation guide, then build your first complete workflow in the quick start.
Documentation map¶
- Getting Started introduces the model, trainer, event, data, and result contracts.
- Guides explain complete user workflows and their trade-offs.
- API Reference is generated from the installed package's public docstrings.
- Examples links the maintained notebooks and their Colab versions.
- Troubleshooting maps common errors to concrete corrections.