Japanese artificial intelligence startup Preferred Networks (PFN) today debuted its PyTorch library pytorch-pfn-extras (PPE). This is the company’s first open-source library supporting research and development in deep learning using the PyTorch framework. The release follows PFN’s December 2019 announcement that it was migrating its deep learning research platform to PyTorch from its own open source deep learning framework, Chainer.
To facilitate the transition, the pytorch-pfn-extras library includes several popular Chainer functions and the following features:
- Extensions and reporter
Functions frequently used when implementing deep learning training programs, such as collecting metrics during training and visualizing training progress
- Automatic inference of parameter sizes
Easier network definitions by automatically inferring the sizes of linear or convolution layer parameters via input sizes
- Distributed snapshots
Reduce the costs of implementing distributed deep learning with automated backup, loading, and generation management of snapshots
PFN is also providing users a thorough Chainer to PyTorch migration guide.
Developed by tech giant Facebook and introduced in October 2016, PyTorch has become one of the most popular open-source deep-learning libraries. In a statement published on the PFN website, the Facebook PyTorch team acknowledged PFN’s ongoing efforts to build and strengthen ties with the PyTorch developer community: “We appreciate PFN for contributing important Chainer functions, such as gathering metrics and managing distributed snapshots, through pytorch-pfn-extras. With this newly available library, PyTorch developers have the ability to understand their model performances and optimize training costs. We look forward to continued collaboration with PFN to bring more contributions to the community, like ChainerRL capabilities later this summer.”
PFN says it is in discussions with the Facebook PyTorch team regarding the possibility of merging PPE’s features into the PyTorch base build. PFN is also planning to open-source a PyTorch version of the deep reinforcement learning library ChainerRL by the end of June 2020.
According to a Nikkei survey, Preferred Networks ranks No.1 on estimated corporate value among 181 Japanese startups, with an estimated valuation of JP¥351.5 billion (US$3.24 billion).
Journalist: Fangyu Cai | Editor: Michael Sarazen
0 comments on “Japanese Unicorn Preferred Networks Releases First PyTorch Library”