OpenAI Fully Embraces Facebook's PyTorch Machine Learning Framework

2020-02-03

OpenAI recently announced that it may migrate its projects to the PyTorch machine learning framework in the future, instead of oscillating between multiple platforms. OpenAI, an AI research company based in San Francisco, was co-founded by CTO Greg Brockman, Chief Scientist Ilya Sutskever, Elon Musk, and others, with support from notable figures such as LinkedIn co-founder Reid Hoffman and former Y Combinator president Sam Altman. In a blog post, the company cited PyTorch's efficiency, scalability, and adoption rate as the reasons for its decision. The company stated in a statement, "Looking ahead, we will primarily use PyTorch as our deep learning framework, but occasionally use other frameworks for specific technical reasons. We are excited to join a rapidly growing developer community, including organizations like Facebook and Microsoft, to drive the scale and performance of graphics cards." OpenAI stated that many of its teams have already migrated their work to PyTorch and will contribute to the PyTorch community in the coming months. Additionally, the company plans to provide Spinning Up in Deep RL educational resources on PyTorch in early 2020. Furthermore, the company intends to research scalable AI systems with data parallel training, visualize the interpretability of these systems, and build a general-purpose robotics framework. (OpenAI is writing PyTorch bindings for its highly optimized blockparse kernel and plans to open-source these bindings in the coming months.) PyTorch was publicly released on Facebook in October 2016 and is an open-source machine learning library based on Torch, a scientific computing framework and scripting language built on the Lua programming language. As of March 2018, it merged with Caffe2, a deep learning toolkit initially developed by researchers at the University of California, Berkeley and further developed by Facebook's AI research lab. Despite TensorFlow being in use for a slightly longer period (since November 2015), PyTorch continues to gain popularity in the data science and developer communities. According to GitHub's 2018 Octoverse report, it became one of the fastest-growing open-source project platforms in the past 12 months. Facebook recently revealed that the number of contributors to the platform increased by over 50% in 2019, reaching nearly 1,200. An analysis conducted by The Gradient found that PyTorch was implemented at every major AI conference in 2019, and O'Reilly noted that PyTorch's citation in papers increased by over 194% in the first half of 2019. Not surprisingly, many leading machine learning software projects have been built on top of PyTorch, including Uber's Pyro and HuggingFace's Transformers. Software developer Preferred Networks recently joined this lineup and pledged to migrate from its custom AI development framework, Chainer, to PyTorch in the near future.

RECENT AI NEWS

RECENT AI TOOLS