PyTorch

06/01/2026

Using PyTorch allowed LinkedIn to enable production-grade optimization at previously infeasible scales.

In our latest PyTorch Foundation case study, the engineering team at LinkedIn shares how they re-architected their extreme scale Linear Programming Solver (DuaLip GPU). Built using PyTorch to enable multi-GPU computations and parallelism, this solver serves as a core building block for massive matching problems that power ranking, personalization, and recommendation systems.

To maximize parallelism, the team tailored GPU ex*****on techniques to sparse matching constraints. They developed constraint-aligned sparse layouts, batched projection kernels, and a distributed design focused on dual variables. They also updated the underlying ridge-regularized dual ascent method with Jacobi-style row normalization, primal scaling, and a regularization parameter continuation scheme.

When tested against extreme scale matching workloads, the GPU implementation achieved a 10x wall-clock speedup over the previous distributed CPU DuaLip solver while maintaining strict convergence guarantees. This project represents an excellent intersection of ML systems, mathematical optimization, and machine learning.

Read the complete case study here 👉 tinyurl.com/mfrup8z2

05/31/2026

One week until the Call for Proposals closes! ⏰

This is your opportunity to join the speaker lineup for PyTorch Conference North America in San Jose (October 20-21). Whether you have a lightning talk idea or a deep-dive session on Responsible AI, we want to see your proposal!

🗓️ Final deadline: June 7
🔗 Apply now: https://bit.ly/4bIgqbs

05/29/2026

Wondering what it's like to attend a PyTorch Conference?

Attendees of the first ever Europe 2026 in Paris share a glimpse into what it's like...

Watch the highlight reel on YouTube: https://youtu.be/0lKdvXZkS-4?si=j53fyUPs_eTAJOqJ

Will we see you at PyTorch Conference China and PyTorch Conference North America later this year?

What was PyTorch Conference Europe 2026, held on April 7–8 in Paris...

05/28/2026

How to Eliminate Pipeline Friction in AI Model Serving

There are numerous issues that are collectively known as pipeline friction, and they cost organizations time, money, and competitive advantage.

This post provides actionable best practices for eliminating the most common sources of friction in AI model serving pipelines. Learn about the most frequent sources of pipeline friction, including unsupported operations, dynamic input sizes, version mismatches, and model export issues that might arise, for example, when converting from a training frameworks like PyTorch into optimized inference formats.

Read the full post:

The path from a trained AI model to production should be smooth, but rarely is. Many teams invest weeks fine-tuning models, only to discover that exporting to a deployment format breaks layers…

05/28/2026

"We believe the future of AI is built on open, production-proven infrastructure - and PyTorch sits at the heart of that future. Joining the PyTorch Foundation is a natural step given our years of running PyTorch at scale across heterogeneous hardware on Alibaba Cloud. We look forward to working alongside the PyTorch Foundation to raise the bar for AI infrastructure and help developers build the next generation of models with confidence," said Dr. Feifei Li, Chief Technology Officer, Alibaba Cloud.

ITBrief covers Alibaba Cloud joining the PyTorch Foundation as a Platinum member. Read the article:

It gives Alibaba Cloud a board seat at the open source group as it seeks to shape AI software used across rival hardware platforms.

05/27/2026

PyTorch Compile can make models run dramatically faster, but the real magic is what happens under the hood.

This blog breaks down one of the most important optimizations behind torch.compile: kernel fusion. Instead of launching separate GPU kernels for every PyTorch operation, PyTorch Inductor can combine dependent operations into a single optimized Triton kernel.

The result? Fewer kernel launches, less memory traffic, fewer intermediate tensors, and more efficient GPU ex*****on.

Learn more here 👉 https://tinyurl.com/ms9sdnyn

05/27/2026

The speed-of-light optimization for Qwen3.5 on the TokenSpeed inference engine is a significant milestone, achieving a record-breaking 580 tokens per second (tps) for agentic workloads on NVIDIA GPUs.

In the PyTorch Foundation's latest community blog post, you can learn all about the complete design, implementation, and optimization of Qwen3.5 models in the TokenSpeed inference framework and see for yourself how this work is improving performance 👉 https://bit.ly/4uGUvIS

This achievement was a joint effort between the Qwen inference team, LightSeek Foundation TokenSpeed team, NVIDIA, and the Mooncake team, with special contributions from Tri Dao for FlashAttention-4 (FA4) optimization.

05/27/2026

Planning your Q4 technical events? Make sure San Jose is on your list! 🏔️

PyTorch Conference North America (Oct 20-21) is the must-attend gathering for anyone serious about the stack ecosystem. Join us for two days of unparalleled education and networking in the center of Silicon Valley.

Early bird registration ends July 31 - don't miss out on $400 in savings.

🔗 Secure your spot: https://bit.ly/4sh3DSw

05/27/2026

We’re excited to welcome Alibaba Cloud as a Platinum Member of the PyTorch Foundation 🎉

Alibaba Cloud is a global leader in full-stack AI infrastructure and the force behind Qwen—one of the world’s most influential open-weight model families. Having run PyTorch at massive scale across diverse hardware, they bring invaluable, production-hardened engineering expertise to the upstream community.

05/27/2026

If you are joining the global community in Shanghai this September 8-9, ensure you register before the July 28 deadline to take advantage of our discounted rates. Attendees can save up to ¥710 RMB (USD$100) on their pass for KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China.

Connect with the maintainers and architects shaping the next generation of digital infrastructure.

🔗 Register today: https://bit.ly/3NG8hv1