neon

Neon is Nervana’s Python-based deep learning library. It provides ease of use while delivering the highest performance. Some of the features that neon has would be the support for commonly used models including convnets, RNNs, LSTMs, and autoencoders, tight integration with neon’s state-of-the-art GPU kernel library, 3s/macrobatch (3072 images) on AlexNet on Titan X (Full run on 1 GPU ~ 32 hrs), basic automatic differentiation support, framework for visualization and swappable hardware backends: write code once and deploy on CPUs, GPUs, or Nervana hardware. Neon supports loading of both common and custom datasets.

Data should be loaded as a python iterator, providing one mini batch of data at a time during training. Users can create their model by providing a list of layers. For layers with weights, provide a function to initialize the weights prior to training. To train a model, provide the training data (as an iterator), cost function, and an optimization algorithm for updating the model’s weights.

To modify the learning rate over the training time, provide a learning schedule. Currently, neon supports Backends, Datasets – such as Images, Text, Video and Custom Datasets; Initializers, Optimizers, Activations, Layers, Costs and Metrics. There are two ways to run models through neon. The first one is to simply execute the python script containing the mode. The second method is to specify the model in a YAML file. YAML is a widely-used markup language.