At PyTorch Conference North America 2026, hear directly from and connect with those working on PyTorch, @vllm_project, @DeepSpeedAI, @raydistributed, Helion, and Safetensors, alongside experts working on GPUs, NPUs, model architectures, inference engines, and architectures spanning the data center and edge. PyTorch Foundation Executive Director @sparkycollier points to why bringing communities together matters: new model architectures, hardware architectures, and inference engines need coordination across software and hardware experts, alongside the open source software being built to support them. Meet the people behind the work October 20–21 in San Jose. Register by September 4 to save on your conference pass: View the schedule: #PyTorchCon#

H3 🐰💋👠 N$FW LoRA -PinkFluffyBunny lora that truly brings the style and pose -Maximum results achieved at 0.5 str on pruned int8 model. -Use the h3_fl2va_pruned_int8_convrot.safetensors model. Alpha quality so temper expectations 👇😉

That’s a wrap on Day 2 keynotes! Visit the team at the PyTorch Foundation booth (G2) in the Solutions Showcase to talk about the work happening across the open source AI stack, from frameworks, compilers, and inference to hardware, infrastructure, and agents. Walk through the Project Pavilion while you’re here. Stop by the @DeepSpeedAI table now until 12:45 pm, then find Safetensors from 1:15 to 3:30 p.m. 🔗 #PyTorchCon# #KubeCon# #CloudNativeCon# #OpenInfraSummit#

At PyTorch Conference North America 2026, hear directly from and connect with the people working on PyTorch, @vllm_project, @DeepSpeedAI, @raydistributed, Helion, and Safetensors, alongside experts working across the AI stack. At #PyTorchCon# NA 2025, Christian Jacobi (@IBM) spoke about clients thinking about deploying AI across core IT and business processes and what comes next: “There’s a lot of innovation ahead of us to do that at scale with the qualities of service that they are used to from classical enterprise computing.” PyTorch conferences are the open source AI community’s town square, where what’s next gets decided. Register by September 4 to save on your conference pass: #PyTorchCon#

At PyTorch Conference North America 2026, hear directly from and connect with the people working on PyTorch, @vllm_project, @DeepSpeedAI, @raydistributed, Helion, and Safetensors, alongside experts working across the AI stack. Keynote speaker @simon_mo_ (@vllm_project, @inferact) says vLLM’s goal is to become “the easiest to use and most efficient inference engine.” At #PyTorchCon# NA 2025, he spoke about the difficulty and cost of using large language models and continuing to push inference efficiency forward. PyTorch conferences are the open source AI community’s town square, where what’s next gets decided. Register by September 4 to save on your conference pass:

We’re proud to announce the inaugural #PyTorchDayJapan#, hosted by PyTorch Foundation, @huggingface, @IBM, and Mitsubishi Electric (@ME_JP_official), on December 10 in Tokyo. #PyTorchDayJapan# will bring together PyTorch enthusiasts, machine learning engineers, AI researchers, and industry professionals for technical talks and discussions spanning training, inference, responsible AI, and more. PyTorch Day Japan is dedicated to open source AI and the impact of PyTorch Foundation projects including PyTorch, @vllm_project, @DeepSpeedAI, Ray, Helion, and Safetensors. 🔗 Learn more:

As Europe’s AI sovereignty debate draws more attention to models, @tech_eu examines PyTorch’s role in the open source infrastructure that supports them. Cate Lawrence (@tech_eu) spoke with Mark Collier, Executive Director, PyTorch Foundation, about PyTorch as a common software layer for open-weight AI, neutral governance for critical AI infrastructure, and PyTorch Foundation’s expanding role as a home for projects including @vllm_project, @DeepSpeedAI, @raydistributed, and Helion. The article also covers SafeTensors moving toward community stewardship under PyTorch Foundation. As Lawrence writes, “For Collier, Europe's competitive advantage isn't simply producing more AI models — it's helping build the open infrastructure that allows an entire ecosystem of companies to innovate.” 🔗 Read it here:

Day 2 is done, and that’s a wrap on #PyTorchConferenceEU# 🇫🇷🔥 A great close to two excellent days at Station F: deep technical content, practical engineering, open collaboration, and an ecosystem with real momentum. This morning I had the chance to deliver the @PyTorch Foundation CTO keynote, focused on ecosystem growth, new working groups, certification, and what it takes to make open source AI easier to build, govern, and scale (OpenMVG). Big Day 2 announcement 🚨 @HuggingFace Safetensors is joining the PyTorch Foundation as a hosted project. That is a meaningful step forward for secure model distribution and trusted AI deployment. (PyTorch) Also great to see keynotes and sessions featuring: • Léonard Hussenot from @GoogleDeepMind's compelling talk on Gemma 4 Edward Yang (@Meta) on PyTorch updates • @LysandreJik (@huggingface) on the Hub as infrastructure and safe, performant model distribution • an outstanding mix of speakers and contributors from @AMD, @RedHat, @NVIDIA, @Google, @IBM, and @LightningAI (PyTorch Conference 2026) Thank you to everyone who made the first PyTorch Conference Europe such a strong debut. The ecosystem is growing. The Foundation is growing. And the community is building what’s next. 🚀 #PyTorch# #OpenSourceAI# #Safetensors# #huggingface#

Nvidia dropped an official DeepSeek-V4.1-Flash NVFP4 build. And this thing is BIG. DeepSeek V4.1 Flash stats 🧠 552B backbone 📚 +196B Engram conditional memory ⚡ only 8B active during prefill 🚀 16B active during decode 👁️ native vision 📖 1 MILLION token context 🧩 384 routed experts across 40 layers 📜 MIT license Nvidia has converted its routed MoE experts to NVFP4 W4A4 specifically for Blackwell GPUs. This is not a compressed giant model down to 4-bit. DeepSeek's experts were already stored in MXFP4, Nvidia instead converts them to its Blackwell-friendly NVFP4 format. And because NVFP4 uses finer scaling, the checkpoint actually gets slightly larger. 💾 Source: ~476 GiB 💾 NVIDIA NVFP4: ~492 GiB 48 safetensor shards. 😳 So why bother? Because NVIDIA is optimizing how those 4-bit experts execute on Blackwell. And impressively, NVIDIA's evaluations show basically no obvious quality collapse from the conversion. For example: 🧠 GPQA Diamond 91.04 → 91.29 💻 SciCode 54.40 → 55.84 🛠️ Terminal-Bench 2.1 81.60 → 82.16 👁️ MMMU-Pro 74.05 → 73.70 Some slightly up. Some slightly down. Essentially benchmark parity. And it already has: ✅ vLLM support ✅ SGLang support ✅ reasoning parser ✅ tool calling ✅ image input ✅ 1M context ⚠️ Nvidia validated it on 4× GB300 GPUs so not a local model (yet). The checkpoint is still ~492 GiB. 🔗 HF: /nvidia/DeepSeek-V4.1-Flash-NVFP4