Yesterday, Baidu’s deep learning open platform PaddlePaddle (PArallel Distributed Deep LEarning), released its latest version with 21 new products such as Paddle Lite 2.0, four end-to-end development kits including ERNIE for semantic understanding (NLP), toolkits and other new upgrades. PaddlePaddle is an easy-to-use, flexible and scalable deep learning platform developed for applying deep learning to many products at Baidu.
Paddle Lite 2.0
The main goal of Paddle Lite is to maintain low latency and high-efficiency of AI applications when they are running on resource-constrained devices. Launched last year, Paddle Lite is customized for inference on mobile, embedded, and IoT devices. It is also compatible with PaddlePaddle and other pre-trained models.
With enhanced usability in Paddle Lite 2.0, developers can deploy ResNet-50 with seven lines of code. The new version has added support for more hardware units such as edge-based FPGA and also permits low-precision inference using operators with the INT8 data type.
New development kits
Development kits aim to continuously reduce the development threshold for low-cost and rapid model constructions.
ERNIE for semantic understanding (NLP): ERNIE (Enhanced Representation through kNowledge IntEgration) is a continual pre-training framework for semantic understanding.
Earlier this year in July, Baidu had open sourced ERNIE 2.0 model and revealed that ERNIE 2.0 outperformed BERT and XLNet in 16 NLP tasks, including English tasks on GLUE benchmarks and several Chinese tasks.
PaddleDetection: It has more than 60 easy-to-use object detection models.
PaddleSeg for computer vision (CV): It is an end-to-end image segmentation library that supports data augmentation, modular design, and end-to-end deployment.
Elastic CTR for recommendation: Elastic CTR is a newly released solution that provides process documentation for distributed training on Kubernetes (k8s) clusters. It also provides the distributed parameter deployment forecasts as a one-click solution.
EasyDL is an AI platform for novice developers to train and build custom models via a drag-and-drop interface. EasyDL Pro is a one-stop AI development platform for algorithm engineers to deploy AI models with fewer lines of code.
The Master mode will help developers customize models for specific tasks. It has a large library of pre-trained models and tools for transfer learning.
Other new upgrades
- New toolkits like graph, federated and multi-task learning.
- API’s upgraded for flexibility, usability, and improved documentation.
- A new PaddlePaddle module for model compression called PaddleSlim is added to enable a quantitative training function and a hardware-based small model search capability.
- Paddle2ONNX and X2Paddle are upgraded for improved conversion of trained models from PaddlePaddle to other frameworks.
Head over to Baidu’s blog for more details.