OpenPPLIt is Shangtang’s open source deep learning reasoning platform based on self-developed high-performance operator library, which can make artificial intelligence applications run efficiently and reliably on existing CPU, GPU and other computing platforms, and provide artificial intelligence reasoning services for cloud scenarios.
At the just held 2021 world Artificial Intelligence Conference (waic), Shangtang technology officially launched the openppl plan, which decided to open up the cloud reasoning ability of deep learning reasoning deployment engine senseppl to the technology community, so as to accelerate the popularization and progress of AI technology!
▎ give reasoning to openppl and time to thinking
Openppl is based on the self-developed high-performance operator library, which has the extremely optimized performance; At the same time, it provides multi back-end deployment capability of AI model in cloud native environment, and supports efficient deployment of deep learning models such as openmmlab.
1、 High performance
The design of micro architecture friendly task / data / instruction multi-level parallel strategy, self-developed NV GPU, x86 CPU computing library, to meet the deployment scenario of neural network reasoning, commonly used image processing performance requirements
- Support fp16 reasoning of GPU T4 platform
- Support CPU X86 platform fp32 reasoning
- Core operator optimization, leading the industry in average performance
2、 Openmmlab deployment
It supports a series of frontier models such as detection, classification, segmentation and super classification in openmmlab, and provides image processing operators for model pre-processing and post-processing
- Follow onnx open standard and provide onnx conversion support
- Support network dynamic characteristics
- Provide high performance implementation of mmcv operator
3、 Multi backend deployment on cloud
For cloud heterogeneous reasoning scenarios, support multi platform deployment
- Support x86 FMA & avx512, NV Turing architecture
- Supporting parallel reasoning of heterogeneous devices
▎ project links
Welcome to star, welcome to issue~
🔗 Contact us: OpenPPL
The development of machine learning is far from over. We will always pay attention to the progress of the industry. Openppl will absorb the needs of the industry, maintain and improve the types of operators and models for a long time, and optimize the whole chain of model reasoning for a long time.
Communication QQ group: 627853444