tango.integrations we provide many ready-to-use component
implementations for leveraging the functionality from popular libraries.
All registered components will be registered under a name that starts with the name of the integration module, possibly followed by a double colon (“::”) and another identifier if there are multiple registered components of a given type.
For example, the
LoadDataset step in the 🤗 Datasets
integration is registered under the name “datasets::load”, and the
TorchFormat format in the PyTorch integration
is registered under the name “torch”.
- 🔥 PyTorch
- 🔥 FairScale
- 🤗 Datasets
- 🤗 Transformers
- ⚖️ Weights & Biases
- 🧪 Beaker
- ☁️ Google Cloud Storage