Hello from the PyTorch Conference in San Francisco! We’re thrilled that we’re able to bring together leading researchers, developers, and academic communities to further the education and advancement of end-to-end machine learning framework.

The PyTorch team has been hard at work this year to bring innovative releases that further enhance the AI and ML community. 

Read on for all of the news and happenings coming out of PyTorch Conference 2023!

Introducing ExecuTorch

ExecuTorch offers a compact runtime with a lightweight operator registry to cover the PyTorch ecosystem of models, and a streamlined path to execute PyTorch programs on edge devices. 

This suite of tools enables ML developers to perform on-device model profiling and better ways of debugging the original PyTorch model.

"ExecuTorch provides a significant update to PyTorch for mobile and edge devices — it will give developers a path which wasn’t available before. I'm pretty excited that it enables a small, high-performing runtime for these devices, and that partners from across the community are authoring delegates to accelerate its programs on their hardware."

Soumith Chintala, Co-Founder of PyTorch

ExecuTorch is architected from the ground up in a composable manner to allow ML developers to make decisions on what components to leverage as well as entry points to extend them if needed. This design provides the following benefits to the ML community:

• Portability: Compatibility with a wide variety of computing platforms, from high-end mobile phones to highly constrained embedded systems and microcontrollers.
• Productivity: Enabling developers to use the same toolchains and SDK from PyTorch model authoring and conversion, to debugging and deployment to a wide variety of platforms, resulting in productivity gains.
• Performance: Providing end users with a seamless and high-performance experience due to a lightweight runtime as well as its ability to utilize full hardware capabilities, including general purpose CPUs and specialized purpose microprocessors such as NPUs and DSPs.
  
  

PyTorch 2.1 Release

At the beginning of the month, we released PyTorch 2.1. PyTorch 2.1 offers automatic dynamic shape support in torch.compile, torch.distributed.checkpoint for saving/loading distributed training jobs on multiple ranks in parallel, and torch.compile support for the NumPy API.

In addition, this release offers numerous performance improvements (e.g. CPU inductor improvements, AVX512 support, scaled-dot-product-attention support) as well as a prototype release of torch.export, a sound full-graph capture mechanism, and torch.export-based quantization.

Along with 2.1, we also released a series of updates to the PyTorch domain libraries. More details can be found in the library updates blog.

Compiling NumPy code into C++ or CUDA via torch.compile

Quansight engineers have implemented support for tracing through NumPy code via torch.compile in PyTorch 2.1.  This feature leverages PyTorch’s compiler to generate efficient fused vectorized code without having to modify your original NumPy code.  It also allows for executing NumPy code on CUDA just by running it through torch.compile under torch.device("cuda").

Huawei and LightningAI Join the PyTorch Foundation as a Premier Members

Huawei and Lightning AI have joined the PyTorch Foundation as premier members.

Huawei works actively on optimizing PyTorch to fully unleash Ascend computing capabilities. Huawei unveiled the All Intelligence strategy to accelerate intelligence across all industries. 

Lightning AI is the company behind PyTorch Lightning, the platform and open-source framework for companies to build and deploy AI products leveraging the latest generative AI models.

We look forward to what they will bring to the foundation for years to come!

Announcing the PyTorch Documentary

The PyTorch Foundation is thrilled to share that we are working on a PyTorch documentary with Speakeasy Productions! The documentary will feature key members and players in the AI and ML space, and include reflections from the birth of PyTorch in 2016, all the way up until now, 2023.

The documentary is scheduled to premiere early 2024, but we’ve released a short trailer! View it here.

Announcing the First PyTorch Contributor Awards

The PyTorch Foundation has started the first annual PyTorch Contributor Awards at the PyTorch Conference.

Each year, individuals will be selected based on their contributions over the year that have enriched the community and significantly accelerated the advancement of AI and machine learning.

This year's winners are:

PY-TORCHBEARER

Excellence in Long-Term Contributions Across All Modalities

• Jack Cao
• Jeff Daily

PYTORCH PACE-SETTER

Excellence in High-Level Activity and Contributions

• James Keeble

PYTORCH NEWCOMER

Excellence in New Contributions

• Tim Dettmers
• Srishti Gureja

PYTORCH AMBASSADOR

Excellence in Bringing New Users to the Community

• Junghwan Park
• Mazen Alotaibi (Ma7dev)
• Stas Bekman

PYTORCH SUPERHERO

Excellence in All Aspects of Community Contributions

• Piotr Bialecki (ptrblck)

PYTORCH REVIEW POWERHOUSE

Excellence in Code Review

• Aaron Gokaslan

PYTORCH PROBLEM SOLVER

Excellence in Uncovering or Resolving Bugs

• Vadim Kantorov

PYTORCH INNOVATOR

Excellence in Innovative New Features or Approaches

• Emcastillo

PYTORCH TRAILBLAZER

Excellence in Documentation and Knowledge Sharing

• José Luis Castro Garcia

PyTorch Training and Certification

We've launched another course with Linux Foundation Training and Certification!

Take an applications-first approach with PyTorch Essentials. Start prototyping AI applications powered by PyTorch by leveraging popular pretrained models in the fields of Computer Vision and Natural Language Processing covering an extensive span of practical applications.

And special for PyTorch Conference this week, take 50% all training courses with code PTCONF23. Valid through October 24.

Tune in

We’re livestreaming the keynotes today, at 9 a.m. PT and 1 p.m. PT. See details and schedule here.

All other sessions will be recorded and uploaded to our PyTorch YouTube channel in the weeks following the conference.

Thank you

Many thanks to our members, partners and community for making PyTorch and this conference a success. We are grateful!

Cheers,

The PyTorch Foundation team