Baidu released a new version of the flying propeller industry-level deep learning open source framework

The U.S. Stock Research Agency learned that according to Titanium Media on December 12, the WAVE SUMMIT+ 2021 Deep Learning Developer Summit hosted by the National Engineering Laboratory for Deep Learning Technology and Applications was held in Shanghai, at which the Flying Propeller Open Source Framework v2.2 was officially released.

Ma Yanjun, senior director of Baidu's deep learning technology platform department, said that as China's first self-developed, feature-rich, open source and open industry-level deep learning platform, Feipao continues to accumulate and break through core technologies, and the newly released open source framework version 2.2 involves deep learning development, training, text task optimization, hardware efficient adaptation, low-threshold reasoning deployment and other innovative technologies, empowering developers and providing technology source power for industrial applications and cutting-edge exploration.

In terms of development, the API of Flying Propeller Framework v2.2 is more abundant, efficient and maintains good compatibility, and more than 100 APIs are enriched in a targeted manner; in terms of training, for different models and hardware, abstracted into a unified distributed computing view and resource view, and through hardware-aware slicing and mapping functions and end-to-end cost models, the optimal model slicing and hardware combination strategies are searched; text tasks are comprehensively accelerated from text processing, training, decoding to deployment. Upgraded support for string tensors to provide developers with an end-to-end text task development experience, and in terms of hardware access, multi-level, low-cost hardware adaptation schemes reduce the cost of adapting the framework and chips.