rmldnn is a command-line tool that makes deep-learning models easy to build and fast to train. It does not require knowledge of any deep-learning framework (Pytorch, Tensorflow, Keras, etc) or any Python code. Using rmldnn, one can build deep-learning models on either a small single GPU laptop or a large supercomputer with hundreds of GPUs or CPUs without any prior knowledge of distributed computing.

Benefits:

• Does not require any knowledge of deep-learning frameworks (Pytorch, Keras, etc) or Python.
• Is designed with scalability & performance in mind from the get-go.
• Is agnostic to machine and processor architectures, running seamlessly on both CPU and GPU systems.
• Runs on anything from one to hundreds of GPUs or CPUs without any required knowledge of distributed computing, automatically managing data partitioning and communication among nodes.
• Allows building models for different computer vision use cases like image classification, object detection, image segmentation, autoencoders, and generative networks.