Baidu has published the PaddlePaddle depth learning toolkit demo.

Chinese search giant Baidu has published a demo version of the source code for its PaddlePaddle (Parallel Distributed Deep LEarning) deep learning tool on GitHub . The announcement of the entire PaddlePaddle code will take place on September 30th.

According to the information on GitHub, the PaddlePaddle assembly is now raw and not ready for mass “use” by interested parties. The developers warn in advance that not all files and packages are ready to be installed yet, so those who want to try out the public demo version can have serious problems when they try to work with PaddlePaddle.

The Chinese note the following features of their development:
')

• flexibility;
• high efficiency and optimization of capacity utilization;
• scalability;
• compatibility with other products.

According to the developers, PaddlePaddle is easier to use than counterparts from Amazon ( DSSTN ), Google ( TensorFlow ) and Microsoft ( CNTK ), which were also published on GitHub for general use.

These tools, according to the Chinese, have a high entry threshold for third-party developers because of their massiveness and versatility. PaddlePadddle is simpler, managed with C ++ or Python. As an illustrative example, the size of the code is given: for solving the same tasks, the code for Paddle will be four times shorter than for competitors.

Every year, more and more functions come to the fore, the work of which can only be provided by intelligent machine tools: voice message translation, image recognition, photo sorting and the correct formation of search results (as the most unattractive function).

At a certain point, technology giants realize that, with all their capital, they are not able to find and hire all talented and enthusiastic people who could work on developing and optimizing DL tools. That is why its source code gets into free access.

» PaddlePaddle on github
» Installation instructions for the demo assembly
» Source Code Documentation

For any problems with the use of PaddlePaddle, developers are asked to contact them directly at paddle-dev@baidu.com .