jrzaurin/pytorch-widedeep

Provides modular deep components for text (RNNs, Transformers via HuggingFace) and images (CNNs) that can be individually trained or combined with tabular embeddings and feature crosses in a unified architecture. Supports multiple tabular encoders including TabMLP, TabResnet, TabNet, and TabTransformer, with flexible preprocessing pipelines for categorical embeddings and continuous feature normalization. Built on PyTorch with a scikit-learn style API through the `Trainer` class for end-to-end training and inference on multimodal datasets.

1,404 stars and 2,740 monthly downloads. No commits in the last 6 months. Available on PyPI.