deepjavalibrary/djl

Supports pluggable deep learning backends (PyTorch, TensorFlow, MXNet) with automatic CPU/GPU selection, enabling seamless engine switching without code changes. Provides a high-level NDArray API and composable neural network blocks for both inference and training, with built-in model zoo integration for pre-trained models. Includes ergonomic dataset handling, training configuration, and optimizer management through a fluent Java API that integrates natively with existing JVM ecosystems.

4,790 stars.