PyTorch Ambassadors are making a big difference, spanning 21 countries and 5 continents. During the #PyTorchCon Europe 2026 Ambassador Meetup, ambassadors met with PyTorch Foundation leadership to share experiences
#Normalization (LayerNorm/RMSNorm) is foundational. Improved #torchcompile on #H100 & #B200 reaches near SOTA kernel speed—17x faster than eager on backwards—with automatic fusion for peak e2e performance.
https://bit.ly/3PXhSyf
Today at #PyTorchCon EU: Talk at 15:40 CEST.
At #PyTorchCon Europe, Matt White, Global CTO of AI and CTO at PyTorch Foundation provides an update on technical strategy, ecosystem and projects and working groups
PyTorch Foundation Announces Safetensors as Newest Contributed Project to Secure AI Model Execution 👏 Safetensors minimizes security risks associated w/ model architectures & execution, providing developers with a trusted path to production. See: pytorch.org/blog/pytorch... #PyTorchCon
#Monarch is the native #PyTorch API for your #Supercomputer.
Update:
-#Kubernetes, EFA, & ROCm support
-Agent-optimized: SQL telemetry & RDMA code sync
-100x smaller install, 8x faster startup
-Distributed AI development that feels like local dev.
https://bit.ly/4bYdMyx
#PyTorchCon
PyTorch Foundation is excited to welcome Helion as a Foundation-hosted project to standardize open, portable, and accessible AI kernel authoring 🎉 #PyTorchCon
Read more: pytorch.org/blog/pytorch...
PyTorchCon EU 2026 Keynote
Kicking off #PyTorchCon Europe with Mark Collier, Executive Director of @pytorch.org Foundation & GM of AI & Infrastructure at @linuxfoundation.org Mark's keynote explores how the open source intelligence stack compounds and what it means for the future of AI infrastructure
Welcome to PyTorch Conference Europe 2026 in Paris! Today's program features keynote sessions from leading voices in AI & deep dives on training, inference, GenAI, and focused tracks on responsible AI & compliance, security & privacy, frameworks & compilers, and more. #PyTorchCon
We’re excited to announce the 2026 PyTorch Docathon May 5-19! Refine technical docs, test tutorials in CI, and accelerate the transition from research to production. Open to all skill levels with support on Discord.
RSVP now: https://bit.ly/4sTVLYb
#PyTorch #OpenSource #AI
CFP is OPEN for PyTorch Conference 2026 in San Jose!
Share advancements in Core PyTorch, vLLM, DeepSpeed, and Ray. Blog: https://bit.ly/4c1ShM1
Deadlines:
Sessions: June 7
Posters: July 26
Save with Super Early Bird rates through April 10.
#PyTorchCon
The PyTorch Ecosystem Working Group welcomes PhysicsNeMo, Unsloth, ONNX, and KTransformers to the Landscape.
This map highlights innovative projects that extend, integrate with, or build upon PyTorch.
Read more: pytorch.org/blog/pytorch-ecosystem-l...
#PyTorch #OpenSource #AI
"PyTorch is probably the most important piece of open source software most enterprise technology leaders have never had a governance conversation about."
Mark Collier at KubeCon on why neutral governance is AI's path to market. Full diginomica.com article: https://bit.ly/4tpIUNa
PyTorch 2.11 Release Live Q&A w/ Andrey Talman & Nikita Shulga on Tuesday, March 31, 10 AM PT.
-Differentiable Collectives
-FlexAttention: FlashAttention-4 on Hopper/Blackwell
-MPS Operator expansion
-RNN/LSTM GPU Export
-XPU Graph
Register: https://pytorch.org/event/pytorch-2-11-release-live-qa/
#NCCL watchdog timeouts are often misunderstood. Meta’s analysis shows >60% are caused by CPU-side stuckness or divergence, not the network. This guide explains using #FlightRecorder to trace collective states and fix hangs
Read: https://bit.ly/4bCqItC #OpenSourceAI #PyTorch
Paris ML Systems Hackathon on April 9
Join #PyTorch Foundation and GPU MODE for a day-long build:
- Distributed training and inference tracks
- B300 and H200 access
- Prizes: GB300 NVL72 rack access
- Talks: PyTorch (Helion), vLLM, Prime Intellect
Register: https://bit.ly/4bSdKqE
PyTorch and Nebius collaborated to speed up DeepSeek-V3 pre-training (16B & 671B) on 256 NVIDIA B200 GPUs. Combining MXFP8 via TorchAO and DeepEP yielded +41% throughput vs BF16.
Full blog:
https://bit.ly/4uN3yIJ
PyTorch 2.11 features improvements for distributed training and hardware operator support. Join Andrey Talman and Nikita Shulga on Tuesday, March 31st at 10 am for a live update and Q&A.
Register: pytorch.org/event/pytorc...
#PyTorch #OpenSource #AI
PyTorch 2.11 is now available, featuring 2,723 commits from 432 contributors. Highlights: FlashAttention-4 for Blackwell/Hopper, Differentiable Collectives, XPU Graph for Intel GPUs, and expanded MPS support.
Release notes: pytorch.org/blog/pytorch...
PyTorch 2.10 is now optimized for Intel Core Ultra Series 3 processors to bring high-performance AI to the PC and edge.
Read our latest blog from the Intel PyTorch and Client AI SW teams for the full technical deep dive and benchmarks:
https://pytorch.org/blog/pytorch-2-10torchao/
TorchSpec and Mooncake teams introduce TorchSpec: a torch-native framework for speculative decoding training. By streaming hidden states via Mooncake, it enables disaggregated pipelines where inference and training scale independently.
https://bit.ly/47eBfIR
Build differentiable computational physics with NVIDIA Warp. It bridges CUDA and Python for high-performance GPU kernels with native auto-diff. Interoperable with PyTorch, JAX, and NumPy.
https://bit.ly/4uG78UQ
GDPA introduces an attention kernel for RecSys, replacing softmax with flexible activations. Deployed in Meta’s GEM model, it achieves 1,145 BF16 TFLOPs (97% utilization) on NVIDIA B200, outperforming FA4 by 3.5× in short K/V settings.
https://bit.ly/418LQl8
#ExecuTorch addresses fragmented native deployment for #AI agents as a #PyTorch native platform. It enables voice models across CPU, GPU, and NPU on Android, iOS, Linux, macOS & Windows
🔗 pytorch.org/blog/building-voice-agen...
Before we head to Paris for PyTorch Conference EU 2026, we’re looking back on 2025 keynotes from visionary AI leaders.
Starting with Eli Uriegas (@_seemethere) from Meta: 11k commits and 794M minutes of CI/CD compute.
Watch: https://youtu.be/xWjXsP1E5mQ?si=JRIVHQ06s3IvYPDq
#PyTorch #OpenSourceAI
MXFP8 training for MoEs on GB200s enables a 1.3x speedup with equivalent convergence versus BF16:
🔗 pytorch.org/blog/mxfp8-training-for-...
PyTorch Foundation is attending Optimized AI Conference in Atlanta, April 14-16. Join 100+ experts to discuss #LLM operations, #RAG, and #InferenceOptimization.
Get 20% off with code: OAIC-20.
Details: oaiconference.com.
#PyTorch #AIInfrastructure #OpenSourceAI
DeepNVMe just got faster and more flexible:
✅ Gen5 NVMe support
✅ 20X faster model checkpointing
✅ Cost-efficient SGLang inference via ZeRO-Inference
✅ CPU-only pinned memory support
📘 pytorch.org/blog/deepnvm...
#PyTorch #DeepSpeed #AIInfrastructure
The #PyTorchFoundation newsletter is your go-to source for the latest updates, events, and community insights to build and innovate with #PyTorch—all in support of accelerating #OpenSourceAI.
📬 Subscribe: pytorch.org/newsletter/
📖 June: pytorch.org/newsletter/j...
Update from the PyTorch ecosystem: The latest NVIDIA
DALI release adds DALI Proxy—making it easier to accelerate parts of your PyTorch DataLoader pipeline without a full refactor.
Learn more
🔗 developer.nvidia.com/blog/unlock-...
#PyTorch #OpenSourceAI #DataPipelines #DeepLearning
