OpenAI releases Triton, a Python-like language for neural-network development

OpenAI has released Triton, an open-source Python-like language for neural-network development.

We’re releasing Triton 1.0, an open-source Python-like programming language for writing efficient GPU code. OpenAI researchers with no GPU programming experience have used Triton to produce kernels that are 2x faster than their PyTorch equivalents. https://t.co/bdovUjTMuF

— OpenAI (@OpenAI) July 28, 2021

The company says Triton will enable maximum hardware performance with minimal effort. With it, developers without CUDA experience can write highly efficient code executed on GPUs.

The language is compatible with Python and has allowed researchers to complete in 25 lines a task that would normally require more effort and resources.

Triton also simplifies the development of specialized kernels that can be far faster than general-purpose library equivalents. The compiler automatically optimizes and parallelizes code, transforming it for execution on Nvidia’s latest GPUs.

“Our researchers have already used [Triton] to create kernels that are twice as efficient as their Torch equivalents. We are excited to work with the community to make GPU programming accessible to everyone,” said Triton creator Philipp Tille.

The first version of the programming language is available on GitHub. To date, it runs only on Nvidia hardware.

Triton was first unveiled in 2019 at the International Seminar on Machine Learning and Programming Languages. Soon its creator Philipp Tille joined OpenAI.

OpenAI disbanded the robotics team in July and halted attempts to apply AI to physical machines.

In June, researchers from the organization OpenAI discovered a way to improve the ‘behavior’ of the GPT-3 language model with respect to ethical, moral and social values.

In March, representatives from OpenAI stated that GPT-3 generates 4.5 billion words daily.

Subscribe to ForkLog News on Telegram: ForkLog AI — all the news from the world of AI!