Baidu Releases PaddlePaddle Upgrades
An updated release of Baidu’s deep learning framework includes a batch of new features ranging from inference capabilities for Internet of Things (IoT) applications to a natural language processing (NLP) framework for Mandarin.
The latest version of PaddlePaddle released this week includes a streamlined toolkit dubbed Paddle Lite 2.0 aimed at inference for IoT, embedded and mobile devices. It works with PaddlePaddle as well as pre-trained models from other sources, Chinese Internet giant (NASDAQ: BIDU) said.
Along with faster deployment of ResNet-50, used for image classification on convolutional neural networks, Paddle Lite 2.0 also supports edge-based FPGAs and other hardware.
New development kits include ERNIE 2.0, and updated version of Baidu’s natural language processing framework. The pre-training framework is said to outperform Google’s BERT technique for NLP training on English and several Chinese language tasks.
As it seeks to attract more machine learning novices, Baidu also said it is adding an AI platform called Easy DL. The tool enables training and building custom models via a drag-and-drop interface. Billed as a “one-stop AI development platform for algorithm engineers to deploy AI models with fewer lines of code,” Easy DL has so far been used by manufacturing, agriculture and service industry users to build more than 169,000 models, Baidu claimed.
Other PaddlePaddle upgrades include expanded support for operators along with easier-to-use APIs. Also included is a “PaddleSlim” module for compressing models.
“By providing hardware support, cloud-to-edge deployment, development kits and Master mode, we’ve significantly improved PaddlePaddle’s performance and feature-set,” said Tian Wu, executive director of Baidu’s AI Group. “PaddlePaddle will keep advancing large-scale distributed computing and heterogeneous computing.
Currently used by more than 1.5 million developers, PaddlePaddle was released to the open-source community in 2016 to help expand AI application development. Baidu Research notes that its platform was used by Intel in developing its Nervana NNP-T processor. Codenamed “Spring Crest,” the Intel neural network processor is designed to train deep learning models at scale and to accommodate a given power budget.