PyTorch

🌟 Join the TorchTPU session at #PyTorchCon Europe 2026 in Paris on April 7–8. Learn the strategy and roadmap that brings a native @PyTorch experience to TPUs. The program includes infrastructure, applications, and agent-based systems, with sessions on training, inference, gen AI, responsible AI, security, privacy, and frameworks.

We're excited to be a community partner for the Optimized AI Conference in Atlanta this week. Today will be our last day at the booth, so stop by and chat with us about the future of open source AI and PyTorch Foundation projects including PyTorch, vLLM, DeepSpeed, and Ray #OAI2026

🗣 Shoutout to our #PyTorchCon Europe gold sponsors. Mark your calendar for 7-8 April in Paris & be part of this historic debut with technical sessions & workshops defining the future of #AI. 🇫🇷 Register: bit.ly/4bUWj91 Schedule: bit.ly/3PpSktm

1️⃣ week until #PyTorchCon Europe! 🇫🇷 Paris becomes the home of #PyTorch for two days of #ML breakthroughs & community from 7-8 April. Check out schedule: bit.ly/3PpSktm. 🎟 Join us! bit.ly/4bUWj91

1 week left to save up to ¥1270! 🎟 Register for #KubeCon + #CloudNativeCon + #OpenInfraSummit + #PyTorchCon China by 7 April for early bird rates. Plus: Got a talk idea? Submit to the #CallForProposals by 3 May! Details: bit.ly/4cduhWr

A round of applause for our #PyTorchCon Europe silver sponsors. Paris becomes the center of the #AI world on 7-8 April. 🌍 Join us for two days packed with world-class keynotes, workshops, & specialized tracks on modern AI development. Register: bit.ly/4bUWj91 Schedule: bit.ly/3PpSktm

The Training Systems track at #PyTorchCon Europe, 7-8 April, explores distributed training, infrastructure optimization & scaling models across modern hardware. Learn more: bit.ly/4uN4bSi 🎟 Register → bit.ly/4bUWj91

PyTorch 2.11 features improvements for distributed training and hardware-specific operator support. Join Andrey Talman and Nikita Shulga on Tuesday, March 31st at 10 am for a live update and Q&A. Topics: - Differentiable Collectives for Distributed Training - FlexAttention: Now includes a FlashAttention-4 backend on Hopper and Blackwell GPUs - MPS (Apple Silicon): Comprehensive operator expansion - RNN/LSTM GPU Export Support - XPU Graph Register: pytorch.org/event/pytorch-… #PyTorch #OpenSource #AIInfrastructure

NCCL watchdog timeouts are some of the most commonly misunderstood and difficult-to-debug errors in AI training. In the past, PyTorch released Flight Recorder to make debugging these errors easier, but interpreting its outputs has been tricky. Based on an analysis of countless NCCL watchdog timeouts at Meta, the PyTorch team at Meta has put together a guide on how to effectively use Flight Recorder to debug NCCL watchdog timeouts, and to better understand why they occur (spoiler: it's usually not because of network or NCCL issues). If you've ever spent hours debugging these errors and are looking for a better solution, this blog post is for you: pytorch.org/blog/flight-re… ✍ Yifei Liu, Uttam Thakore, Ph.D., Junjie W., Yongzhong Yang #PyTorch #OpenSourceAI #NCCL #FlightRecorder

GPU MODE (@GPU_MODE) & PyTorch Foundation are organizing an ML systems hackathon in Paris on April 9, immediately following PyTorch Conference Europe 2026. Researchers and engineers will compete across two tracks: -Distributed training (LLM speedrun) and inference optimization (leaderboard) - Access to a B300 cluster from Verda and H200 instances from Sesterce - Cloud credits as prizes, including 48-hour access to a GB300 NVL72 rack - Talks from PyTorch (Helion), vLLM, Prime Intellect, and more - Food and refreshments Doors open at 9:30, with the closing ceremony at 20:00. Attendees can join with a pre-formed team or match on-site. Location details are shared upon registration. Spots are limited. Register: luma.com/gpu-mode-paris… #OpenSourceAI #PyTorchCon #PyTorch #vLLM #Helion

🔐 #AI systems must be secure by design. The Security & Privacy track at #PyTorchCon Europe (7-8 April, Paris) explores protecting models, safeguarding data & defending AI systems from emerging threats. Learn more: bit.ly/47fbY1o 🎟 Register → bit.ly/4bUWj91

PyTorch and Nebius have collaborated to speed-up pre-training DeepSeek-V3 (16B & 671B) on 256 NVIDIA B200 GPUs using TorchTitan. Combining MXFP8 via TorchAO with DeepEP yielded up to 41% faster training throughput over BF16 (with equivalent convergence). Full results and reproducible configs in the blog: 🔗 pytorch.org/blog/enabling-… ✍ Alireza Shamsoshoara, Matthias Reso, Hamid Shojanazeri, @vega_myhre, Hooman Ramezani #PyTorch #OpenSourceAI #TorachAO #MXFP8

⚖️ Building #AI responsibly matters. The Responsible AI & Compliance track at #PyTorchCon Europe, 7-8 April in Paris, explores governance, fairness, transparency & regulatory considerations in AI development. Learn more: bit.ly/40PqoBz 🎟 Join us → bit.ly/4bUWj91

Ready to take the stage in Shanghai? 🎤 The #CallForProposals is taking submissions for #KubeCon + #CloudNativeCon + #OpenInfraSummit + #PyTorchCon China, 8-9 September! Share your technical breakthroughs in #CloudNative, #AI, or infrastructure. 🗓 Submit by 3 May: bit.ly/40QIjb2

Meta, in collaboration with @huggingface and PyTorch Foundation, is launching India’s first OpenEnv AI Hackathon. Participants will build reinforcement learning environments using OpenEnv, contributing directly to how intelligent agents are trained. @Meta What's in it for developers? $30,000 prize pool! Round 1 (Online): March 25 – April 8 Finale (Bangalore): April 25 – 26 (48-hour in-person hackathon at Scaler School of Technology) Learning resources are provided for those without prior RL experience. 🔗 Register by April 3: pytorch.org/event/openenv-… #PyTorch #OpenEnv #MetaAI #HuggingFace #OpenSource

We have to get out of our silos and tribes that deliver just pieces for users to figure out, and actually do co-design and co-evolve the full stack to deliver what the world needs.

Today at KubeCon + CloudNativeCon Europe during keynote fireside chat @sparkycollier, PyTorch Foundation ED, explained that the problems we need to solve to deliver open source AI for training, inference and agents require deep expertise across different domains, and those experts work in communities organized by a number of different foundations. To train and run thousands of models, @Uber relies on PyTorch Foundation projects: @PyTorch, @raydistributed & @vllm_project and runs on @CloudNativeFdn project @kubernetesio and the newest CNCF project, llm-d, is tightly integrated with vLLM #KubeCon #CloudNativeCon #PyTorch #Ray #vLLM #Kubernetes #CloudNative #OpenSource

PyTorch 2.11 features improvements for distributed training and hardware-specific operator support. Join Andrey Talman and Nikita Shulga on Tuesday, March 31st at 10 am for a live update and Q&A. Topics: - Differentiable Collectives for Distributed Training - FlexAttention: Now includes a FlashAttention-4 backend on Hopper and Blackwell GPUs - MPS (Apple Silicon): Comprehensive operator expansion - RNN/LSTM GPU Export Support - XPU Graph Register: pytorch.org/event/pytorch-… #PyTorch #OpenSource #AIInfrastructure