Thriving 
JV A lightweight Java deep learning library for building, training, and running neural networks, with optional Apple Metal acceleration.
https://kirstenali.github.io/DeepJ/landing ↗ HEALTH ? 100 / Healthy
Thriving JV // readme
DeepJ
DeepJ is a Java library for building, training, and running neural networks. It provides N-dimensional tensors, automatic differentiation, composable layers, and CPU execution with optional Apple Metal acceleration.
Build feed-forward networks, convolutional networks, recurrent networks, autoencoders, and Transformers on the same tensor and autograd foundation.
Website · Documentation · API reference
Supported networks
| Model type | Implementations |
|---|---|
| Feed-forward networks | FNN, with configurable hidden and output activations |
| Convolutional networks | Conv1D, Conv2D, max/average pooling, and global average pooling |
| Recurrent networks | RNN, GRU, LSTM, individual cells, explicit state, and per-sample sequence lengths |
| Autoencoders | Composable Autoencoder with separate encoder and decoder access |
| Transformer encoders | Bidirectional attention, padding masks, and masked mean pooling |
| Decoder Transformers | DeepJ Origin, Orbit, Prism, Nova, Loop, Qwen3, and Gemma 2 for causal… |