Nominations Open for the 2026 PyTorch Foundation Contributor Awards
AnnouncementsBlog
Nominations are now open for the 2026 PyTorch Foundation Contributor Awards! These awards recognize outstanding…
PyTorch Meetup Singapore: A milestone in APAC
Blog
TL;DR Eighty engineers, researchers, and community builders gathered for the inaugural PyTorch Meetup Singapore. Hosted…
Portable vLLM Model Inference Kernels in Helion
Blog
TL;DR Helion kernels were integrated into vLLM for FP8 inference using Qwen3 models and evaluated…
Using Muon Optimizer with DeepSpeed
Blog
TL;DR DeepSpeed now supports Muon Optimizer! Muon Optimizer has gained great momentum with significant adoption…
How LinkedIn Uses PyTorch to Solve Extreme-Scale Optimization Problems
BlogCase Studies
TL;DR: This case study demonstrates how LinkedIn re-architected its distributed linear programming solver, DuaLip, by…
Why Is PyTorch Compile So Fast: Kernel Fusion
Blog
When you use PyTorch's compiler, your model runs faster, up to 10x faster. But what's…
Up to 580tps! New Speed Record of Qwen3.5-397B-A17B on GPU for Agentic Workloads with TokenSpeed
Blog
TL;DR: The TokenSpeed inference engine achieved a record-breaking 580 tps running the Qwen3.5-397B-A17B model on…
Alibaba Cloud Joins the PyTorch Foundation as a Platinum Member
AnnouncementsBlog
The PyTorch Foundation, a community-driven hub for open source AI under the Linux Foundation, is…
TLX Block Attention: A Warp-Specialized Blackwell Kernel for Fixed-Block Sparse Self-Attention
Blog
Code available at: https://github.com/facebookresearch/ads_model_kernel_library In this post, we present the design of TLX Block Attention…
Join the PyTorch Foundation Ambassador Program: A Global Network of Community Leaders
AnnouncementsBlog
A little over a year ago, the PyTorch Foundation launched the Ambassador Program, an initiative…
PyTorch Docathon 2026 Results in 150+ Merged Pull Requests
AnnouncementsBlog
Thank you to everyone who participated in the PyTorch Docathon 2026! Once again, the community…
vLLM and PyTorch Work Together to Improve the Developer Experience on aarch64
Blog
TLDR: PyTorch 2.11 makes it possible to install CUDA-enabled PyTorch wheels on aarch64 Linux directly…
Running PyTorch Models on Apple Silicon GPUs with the ExecuTorch MLX Delegate
Blog
TL;DR: Introducing the ExecuTorch MLX Delegate The new MLX delegate enables optimized, GPU-accelerated inference for…
PyTorch 2.12 Release Blog
Blog
We are excited to announce the release of PyTorch® 2.12 (release notes)! The PyTorch 2.12…
Efficient Edge AI on Arm CPUs and NPUs: Understanding ExecuTorch through Practical Labs
Blog
TL;DR: ExecuTorch extends the PyTorch ecosystem to deliver local AI inference on constrained edge devices.…
In-Kernel Broadcast Optimization: Co-Designing Kernels for RecSys Inference
Blog
TL;DR: Traditional RecSys inference explicitly replicates shared user embeddings/sequences for every candidate. In-Kernel Broadcast Optimization…
SMG: The Case for Disaggregating CPU from GPU in LLM Serving
Blog
How It Started: Hitting the GIL Wall at Scale We've been running production model serving…
Introducing AutoSP
Blog
¹ SSAIL Lab, University of Illinois Urbana-Champaign, ² Anyscale, ³ Snowflake TL;DR: AutoSP automatically converts…
IBM Research uses vLLM at the heart of its RITS Platform
BlogCase Studies
TL;DR: vLLM has been critical to democratizing access to our research community to the latest…
Optimizing Effective Training Time for Meta’s Internal Recommendation/Ranking Workloads
Blog
Motivation and Introduction Across the industry, teams training and serving large AI models face aggressive…