PyTorch Conference 2025 Recap
PyTorch Conference 2025 brought together 3,432 developers, researchers, and innovators from 1,026 organizations across the global AI ecosystem for two days of keynotes, technical sessions, and community connection. The program featured 22 keynotes, 103 sessions, and 113 poster presentations, reflecting the scale and diversity of the PyTorch community.
Read our PyTorch Conference recap: Accelerating the Future of Open Source AI
View the Recordings
You can now view all recordings from the conference and its co-located events on the PyTorch YouTube Channel:
PyTorch Conference 2025 delivered major announcements like Ray joining PyTorch Foundation to help advance a unified distributed compute stack, Dell Technologies joined as a Premier Member, and Qualcomm joined as a Premier Member.
The Meta PyTorch team introduced several new PyTorch-native technologies and accompanying blogs:
- Helion: A High-Level DSL for Performant and Portable ML Kernels
- Introducing PyTorch Monarch
- Monarch + Lightning AI: Unlocking New Possibilities in Distributed Training
- torchcomms: A Modern PyTorch Communications API
- Introducing torchforge – a PyTorch Native Library for Scalable RL Post-Training and Agentic Development
- Introducing ExecuTorch 1.0: Powering the Next Generation of Edge AI.
Ecosystem News
The PyTorch Ecosystem continues to expand with LMCache and DeepInverse, both advancing open source tools for inference and imaging.
PyTorch Contributor Awards
The Foundation also celebrated its 2025 PyTorch Contributor Awardees, recognizing community members who have made outstanding contributions across development, research, and advocacy.
Congratulations to Donny Greenberg and Runhouse on winning the Startup Showcase!
PyTorch 2.9 Now Available
Read the release blog
Inside Helion: Live Q&A with the Developers
December 11, 11 am PST. Register today!
Member Round-Up
PyTorch Team at Meta: KernelFalcon: Autonomous GPU Kernel Generation via Deep Agents
vLLM Team at IBM: Hybrid Models as First-Class Citizens in vLLM
vLLM Team at IBM Research, vLLM Team at Red Hat, and vLLM Team at AMD: Enabling vLLM V1 on AMD GPUs With Triton
Snowflake: SuperOffload: Unleashing the Power of Large-Scale LLM Training on Superchips
PyTorch Team at Meta: When Quantization Isn’t Enough: Why 2:4 Sparsity Matters
Community Events
- AI By the Bay: Nov 17–19, 2025, Oakland, CA
- NeurIPS: Dec 2–7, 2025, San Diego, CA
- Open Source Summit Japan: Dec 8–10, Tokyo, Japan
- Inside Helion: Live Q&A with the Developers: Dec 11, Virtual
Subscribe to the PyTorch Newsletter
Get updates like these delivered directly to your inbox. Subscribe to the PyTorch newsletter for announcements, event highlights, technical deep dives, and more: https://pytorch.org/newsletter/