scorcher

Modular deep learning in R with torch, designed for prediction-powered inference workflows.

scorcher provides a composable, Keras Functional-style interface for building deep learning models in R using torch, with special attention to workflows needed for inference with predicted data.

Highlights:

• Functional API for multi-input, multi-output architectures
• Helpers for Hugging Face models and multimodal pipelines
• Designed to play nicely with ipd and related inference tooling

• GitHub: https://github.com/jtleek/scorcher
• Documentation: in development