Ray Train: Scalable Model Training
Contents
Ray Train: Scalable Model Training#
Tip
Train is currently in beta. Fill out this short form to get involved with Train development!
Ray Train scales model training for popular ML frameworks such as Torch, XGBoost, TensorFlow, and more. It seamlessly integrates with other Ray libraries such as Tune and Predictors:
Intro to Ray Train#
Framework support: Train abstracts away the complexity of scaling up training for common machine learning frameworks such as XGBoost, Pytorch, and Tensorflow. There are three broad categories of Trainers that Train offers:
Deep Learning Trainers (PyTorch, TensorFlow, Horovod)
Tree-based Trainers (XGboost, LightGBM)
Other ML frameworks (HuggingFace, Scikit-Learn, RLlib)
Built for ML practitioners: Train supports standard ML tools and features that practitioners love:
Callbacks for early stopping
Checkpointing
Integration with TensorBoard, Weights/Biases, and MLflow
Jupyter notebooks
Batteries included: Train seamlessly operates in the Ray ecosystem.
Use Ray Data with Train to load and process datasets both small and large.
Use Ray Tune with Train to sweep parameter grids and leverage cutting edge hyperparameter search algorithms.
Leverage the Ray cluster launcher to launch autoscaling or spot instance clusters on any cloud.
Next steps#
