In February 2019, Uber open sourced Ludwig to help users with less engineering and machine learning expertise develop, train, and test deep learning models for use cases ranging from natural language understanding to time series forecasting. More recently, Uber announced Ludwig version 0.3, which was developed in collaboration with Stanford’s HazyResearch group. The new version features a modular and extensible backend based on Tensorflow 2; which enables integration with Hugging Face’s Transformers repository and Weights and Biases experiment management platform, and support for new data formats (Apache Parquet, JSON, TSV, JSONL). It also includes a command to perform automated hyperparameter optimization, k-fold cross validation capabilities, and a vector data type for model training with weak supervision.