PyTorch

MXFP8 training for MoEs on GB200s enables a 1.3x speedup with equivalent convergence versus BF16:

🔗 https://pytorch.org/blog/mxfp8-training-for-moes-1-3x-training-speedup-vs-bf16-for-llama4-scout-on-gb200-cluster-using-torchao-and-torchtitan/

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...

Unlock Efficient Data Processing with the Latest from NVIDIA DALI | NVIDIA Technical Blog NVIDIA DALI, a portable, open source software library for decoding and augmenting images, videos, and speech, recently introduced several features that improve performance and enable DALI with new use...

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

🧠 Responsible AI is a design decision—and a strategic edge.

This new guide shows how to build a Yellow Teaming assistant using PyTorch and AWS Graviton4 to surface risks early and build more accountable systems.

🔗 pytorch.org/blog/build-r...

#ResponsibleAI #LLM #PyTorch #builtonArm

⏳ Just a few days left to apply for the PyTorch Ambassador Program.

If you're making an impact with PyTorch through research, code, education, or community work, now’s your chance to join a global network of ML leaders.

📅 Deadline: June 7
🔗 pytorch.org/programs/amb...
#PyTorch #AICommunity

NVIDIA GTC Paris 2025 - Schedule Browse technical sessions designed to sharpen your skills and expand your expertise.

Join us at #GTC25Paris25 for the session “10x Your GPU Power with #Python: Python for Programming the GPU”

Learn how Python now matches the performance and control of C++ #CUDA.

Explore #PyTorch, CuPy, RAPIDS, cuda.parallel, numba.cuda, cuTile, etc.
🔗 www.nvidia.com/en-eu/gtc/se...

Mixture-of-Experts (MoE) is a popular #LLM architecture that reduces computation by activating fewer parameters per token. But it brings memory, communication, & control challenges.

💡We introduce MetaShuffling, enabling efficient Llama 4 model inference in production. 🔗 pytorch.org/blog/metashu...

The PyTorch Foundation is a Gold Sponsor of #MLSys2025 this week in Santa Clara.

Visit the booth and explore talks from Soumith Chintala, Ion Stoica, and Exec Dir Matt White on open source AI and scalable ML systems.

🔗 pytorch.org/blog/pytorch...

#PyTorch #OpenSourceAI #AIInfrastructure

🖇️: pytorch.org/event/toward...

PyTorch Foundation has expanded into an umbrella foundation.
vLLM and DeepSpeed have been accepted as hosted projects, advancing community-driven AI across the full lifecycle.

Quotes from AMD, AWS, Arm, Huawei, HuggingFace, IBM, Intel, LightningAI, Meta.

Read more: pytorch.org/blog/press-r...

Can language model systems autonomously complete entire tasks end-to-end?

In our next Expert Exchange webinar, Ofir Press explores autonomous LM systems for software engineering, featuring SWE-bench & SWE-agent—used by OpenAI, Meta, & more.

🔗 pytorch.org/autonomous-l...

#PyTorch #AI #OpenSource

PyTorch 2.7 Release Live Q&A

TODAY: Join PyTorch Core Maintainers Piotr Bialecki (NVIDIA) and Nikita Shulga (Meta) for a live Q&A session on the #PyTorch 2.7 release at 12 PM PST.

Have questions? Drop them below, & we'll share them during the webinar.

🔗 More info: pytorch.org/pt-27-releas...

#MachineLearning #OpenSourceAI

Update from the PyTorch maintainers: 2.7 is out now.
🔹 Support for NVIDIA Blackwell (CUDA 12.8)
🔹 Mega Cache
🔹 torch.compile for Function Modes
🔹 FlexAttention updates
🔹 Intel GPU perf boost
🔗 Blog: hubs.la/Q03jBPSL0
📄 Release notes: hubs.la/Q03jBPlW0
#PyTorch #OpenSourceAI

The PyTorch Day France 2025 schedule is now live:
Explore the full agenda of talks and sessions
⛓️ pytorchdayfrance2025.sched.com
Co-located with #GOSIMAI2025

🎟️ Use code PYTORCHFRIEND for 25% off registration
👉 Or enter the Lucky Draw: paris2025.gosim.org

#PyTorch #PyTorchDayFrance

🔊 PyTorch's updated Sphinx theme is now in the main branch on docs.pytorch.org (coming to stable in v2.8)!
This update features dark mode, page ratings, expandable nav & more.
Try it out and share feedback via our survey: forms.gle/VJCypjGdZ1Ty.... #PyTorch #Documentation

GOSIM AI Paris 2025 GOSIM Paris, the City of Light, transforms into the City of Artificial Brilliance this May. GOSIM AI 2025 invites visionaries, disruptors, and pioneers to converge at Station F, a crucible of innovati...

Enter GOSIM Foundation's Lucky Draw for 70–90% off PyTorch Day France tickets—co-located with GOSIM AI Paris 2025.

👉 Look for banner at paris2025.gosim.org

Schedule: paris2025.gosim.org/schedule-day...
Info: events.linuxfoundation.org/pytorch-day-...

#PyTorch #GOSIMAIParis #PyTorchDayFrance

PyTorch 2.7 Release Live Q&A

Curious about what’s coming in PyTorch 2.7?

Core Maintainers Piotr Bialecki (NVIDIA) and Nikita Shulga (Meta) will take them live during a Q&A on April 28 at 12 PM PST.

Hear directly from the folks behind CUDA, CI, and releases.

🔗 pytorch.org/pt-27-releas...

#PyTorch #PyTorch27 #OpenSourceAI #ML

🚀 The EU AI Act is here—the world’s first comprehensive AI regulation. While it recognizes open source AI’s value, its exemptions aren’t unlimited.

🔹 Who’s affected
🔹 Open source exemptions
🔹 GPAI provider obligations

linuxfoundation.eu/newsroom/ai-...

#AIAct #OpenSourceAI

DeepSeek Moments Fuel PyTorch and Open Source Over the Long Haul How open source PyTorch helps DeepSeek make AI performance breakthroughs and unlocks AI innovation for all Chinese AI startup DeepSeek followed up its epic R1 launch with a Week of Open Source, a fest...

Jim Zemlin on how open source PyTorch powers DeepSeek's AI breakthroughs and expands access to innovation: lnkd.in/earz5jQa

DeepSeek is building an F1 racer—fast and specialized. PyTorch is an all-terrain vehicle—modular and open for anyone to customize their ML stack.

#OpenSource #AI #PyTorch

PyTorch Site Survey We appreciate the time you've taken to provide your feedback. This survey aims to help the PyTorch team better understand how users interact with the PyTorch Docs and Tutorials site, enabling us to m...

Please fill out the linked form to participate in our documentation survey to help the PyTorch documentation team know which areas to focus on to improve your docs experience: forms.gle/KZ4xGL65VRMY...

PyTorch documentation is the cornerstone for how developers get the information they need about PyTorch!

As such, the PyTorch documentation team is looking towards improving this overall experience and would love your feedback on how we can improve! ⏬

Using PyTorch and DINOv2 for Multi-label Plant Species Classification

Explore how PyTorch and DINOv2 power multi-label plant species classification in our upcoming webinar with Intel's Murilo Gustineli on March 27 at 12 PM PST.

👋 Register today: pytorch.org/pt-dinov2-mu...

#pytorch #machinelearning #optimization

Hackathon 2025 SemiAnalysis is kicking things off ahead of NVIDIA GTC! Start your day with engaging morning keynotes, hack all day with low-level NVIDIA GPU programming (maybe even Blackwell), take a breather wit…

We're sponsoring TODAY's SemiAnalysis GPU Hackathon in San Jose ahead of GTC 🎙️ Speakers: Mark Saroufim, Vijay Thakkar, Horace He, Philippe Tillet & Tri Dao 🏆 Prizes include hundreds of GPU compute credits for top participants. More: semianalysis.com/hackathon-20... We can't wait to see you there!

Join us in San Francisco Oct 22-23 to showcase your expertise at #PyTorchConf 2025! Share insights with the global #AI community at this industry-leading #OpenSource #ML framework event. Submit proposals for sessions, lightning talks & more by June 1: hubs.ly/Q03bpZ310

Peak Performance, Minimized Memory: Optimizing torchtune’s performance with torch.compile & Liger Kernel LinkedIn: Shivam Sahni, Byron Hsu, Yanning Chen Meta: Ankith Gunapal, Evan Smothers

Explore the integration of a custom #triton kernel, Liger Kernel w/ torch.compile to enhance the performance of fine-tuning #LLMs using #torchtune.

💡 Results show a 47% reduction in peak GPU memory allocation at batch size 256 with meta-llama/Llama-3.2-1B

📖 Read more: pytorch.org/blog/peak-pe...