The Livermore Big Artificial Neural Network toolkit (LBANN) is an open-source, HPC-centric, deep learning training framework that is optimized to compose multiple levels of parallelism.
v0.9 is LBANN’s initial release. The release notes linked below contain details about v0.9x iterations. Some deprecated features have been retired. At a high level, this release includes new development in these areas:
- Support for new training algorithms (e.g., generative adversarial networks)
- Support for new network structures
- Support for new layers (e.g., learning, metrics, optimizers, activations)
- Performance optimizations (e.g., GPU model-parallel softmax layer)
- Model portability & usability
- Support for an online, distributed data store
- Overhauled I/O & data readers
- Overhauled code base (see README_coding_style.txt)
- New features & build system (e.g., support for convolutional and pooling layers, GPU acceleration of local Elemental GEMM operations)
- and much more!