PyTorch Conference 2026 Set for San Jose in October

The Linux Foundation will host PyTorch Conference North America 2026 on October 20 and 21 at the San Jose McEnery Convention Center in California, bringing together the deep learning framework's core contributors and infrastructure partners for two days of technical sessions.

The event is organized under the Linux Foundation banner, which stewards the PyTorch Foundation, and gathers keynote speakers from Meta, Google Cloud, Red Hat, Nvidia, Cohere, the vLLM project, and the Agentic AI Foundation. That lineup spans the major hardware, cloud, and open-source players shaping how large models get trained and served today.

Who it's for

This conference targets ML engineers, systems researchers, and infrastructure teams working directly with PyTorch in production or research settings. It is less a general AI industry showcase than a working conference for people optimizing training pipelines and debugging model internals.

What's on the agenda

Confirmed sessions cover CUDAGraph observability, TorchDynamo-based debugging, and multi-node foundation model training, topics aimed squarely at practitioners scaling models across GPU clusters. These are the kinds of low-level performance and tooling issues that rarely get airtime at broader AI conferences but matter enormously to teams running large-scale training jobs.

The inclusion of the vLLM project and the Agentic AI Foundation among keynote contributors signals an agenda that stretches beyond core framework updates into inference serving and agentic system design, reflecting how PyTorch's ecosystem has broadened well past traditional model training.

When and where to register

The conference runs October 20-21, 2026, at 150 W San Carlos St in downtown San Jose. No registration link has been published yet in the source listing, so attendees should watch the official PyTorch Foundation and Linux Foundation event channels for ticketing details as the date approaches.