|AWS And Facebook Launch PyTorch Tools|
|Written by Alex Denham|
|Friday, 05 June 2020|
Two new tools have been released for PyTorch, the open source library for deep learning. Both are collaborations between Amazon AWS and Facebook. TorchServe is a PyTorch model serving library, while the TorchElastic Controller for Kubernetes adds Kubernetes support to TorchElastic, a library for fault-tolerant and elastic training in PyTorch.
PyTorch is an optimized tensor library for deep learning using GPUs and CPUs. It aims to offer a replacement for NumPy that make suse of the power of GPUs, while providing a deep learning research platform that provides maximum flexibility and speed.
TorchServe aims to provide a clean, well supported, and industrial-grade path to deploying PyTorch models for inference at scale without having to write custom code. TorchServe provides a low latency prediction API, and also embeds default handlers for the most common applications such as object detection and text classification. It also includes multi-model serving, model versioning for A/B testing, monitoring metrics, and RESTful endpoints for application integration.
The Kubernetes Controller with TorchElastic integration gives PyTorch developers a way to train machine learning models on a cluster of compute nodes that can dynamically change without disrupting the model training process. If a node goes down, TorchElastic can pause node level training and resume once the node is healthy again. By using the Kubernetes controller with TorchElastic, distributed training jobs can be run on clusters with nodes that get replaced, either due to hardware issues or node reclamation. This means developers can create training systems that can work on large distributed Kubernetes clusters that include cheaper spot instances. Such instances can vary significantly depending on how many unused EC2 instances are available, and are liable to interruption, which would cause problems with traditional machine learning training frameworks.
or email your comment to: email@example.com