Don't miss @DeepSpeedAI virtual office hours on May 26 at 12:00 PM America/New_York to ask questions of @toh_tana member of DeepSpeed TSC & get the latest recent key updates, including AutoSP (sequence parallel), AutoEP (expert parallel), and AutoTP (tensor parallel).

DeepSpeed の新機能 AutoSP のPyTorch公式ブログが公開されました！ - コンパイラレベルでの最適化により、既存モデルに設定変更だけで Sequence Parallel を適用 - 長い系列の学習に最適化された Sequence-aware AC (activation checkpointing) これにより、長い系列の学習を、より高いGPU効率で容易に実現できます。 pytorch.org/blog/introducing…

Great News! Thanks to DeepSpeed AutoSP, efficient long context LLM training is now easily accessible.

DeepSpeedのAutoSP (sequence parallel) という機能が、PyTorchの公式ブログに載りました！ 論文は、今週開催されていたICLRにも採択されています。 pytorch.org/blog/introducing…

Really happy to see our work, AutoSP, featured as a PyTorch Foundation Blog! AutoSP automatically converts existing single-GPU training pipelines to memory-efficient Sequence-Parallel code purely through a set of compiler passes. Users now no longer need to change user-level code libraries to implement long-context distributed training strategies, a huge pain point for teams! Check out our work: Project page: supercomputing-system-ai-lab… Code: github.com/deepspeedai/DeepS… Paper: supercomputing-system-ai-lab…

The AutoSP project just went live on the PyTorch blog. AutoSP is a compiler-based system for long-context LLM training. It automatically applies sequence parallelism and long-context-aware checkpointing, enabling longer context training without reworking the system stack. Great collaboration with an excellent team: Ahan Gupta, Zhihao Wang, Neel Dani, Masahiro Tanaka, Tunji Ruwase Code and tutorial are fully open-sourced: github.com/deepspeedai/DeepS… Blog: pytorch.org/blog/introducing… Project page: supercomputing-system-ai-lab… Paper: supercomputing-system-ai-lab…

Want to train LLMs on longer contexts without re-engineering your entire systems stack? Introducing AutoSP — the first compiler-based solution that automatically optimizes LLM training for long contexts. Under the hood, AutoSP applies a series of compiler passes that trigger sequence parallelism, paired with a curated activation-checkpointing scheme tailored for long-context training. It's integrated directly into DeepSpeed, so enabling long-context training is just a config change away. No more rewiring your stack to push context lengths. Read the blog to learn more 🖇️ pytorch.org/blog/introducing… ✍ @AhanGupta13, Zhihao W., Neel Dani, @toh_tana, Tunji Ruwase, @_Minjia_Zhang_ #PyTorch #DeepSpeed #AutoSP #OpenSourceAI

1/🚨 A bit late to the party, but excited to share our paper at #ICLR2026: AutoSP — the first compiler-based solution to unlock long-context LLM training. ✅ Up to 2.7× longer contexts on NVIDIA (2.5x on AMD) ✅ Negligible runtime overhead ✅ Merged into DeepSpeed 🚀 🧵👇

Grateful to share that SSAIL lab had 4 papers accepted to ICLR 2026 and 1 paper accepted to MLSys 2026, with all results released on the same day! The accepted works span long-context extension, automated sequence parallelism, lookahead GRPO optimization, agentic multimodal tool using, and high-performance LLM inference. Huge credit to the students and collaborators for pushing these ideas through many iterations, and thanks to the reviewers for the thoughtful feedback. ICLR'26 1. From Collapse to Control: Understanding and Extending Context Length in Emerging Hybrid Models via Universal Position Interpolation, openreview.net/pdf?id=MjmORK…, fantastic work by Haochen Shen and Zheng Wang, in collaboration with IBM researchers Davis Wertheimer, Naigang Wang, Mudhakar Srivatsa, Raghu Ganti 2. AutoSP: Unlocking Long-Context LLM Training Via Compiler-Based Sequence Parallelism, openreview.net/pdf?id=0fgsHv…, led by Ahan Gupta, with Zhihao Wang, Neel Dani, and collaborators Tunji Ruwase and Masahiro Tanaka 3. Slow-Fast Policy Optimization: Reposition-Before-Update for LLM Reasoning, openreview.net/pdf?id=xBlHiH…, Co-led by Zheng Wang and the team. 4. VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use, openreview.net/pdf?id=Idst6X…, led by Mingyuan Wu. MLSys'26 5. SuperInfer: SLO-Aware Rotary Scheduling and Memory Management for LLM Inference on Superchips, led by Jiahuan Yu with Mingtao Hu and Zichao Lin. #SSAIL #MLSys #ICLR

the issue with the summer break is rags like autosp*rt need more clicks so they’ll just post any old shite to get engagement

❓ ciao uolfiiii ti viene in mente per caso qualche ff ambientata in estate au o normale? grazie 💞💞💞💞 💬 Celeste nostalgia di actionratio, Ancora cinque minuti di Ilyfraus, Voce del mare di mityboh, Golden hour di Lyssnokomis (e mi autosp... tellonym.me/uolfinaa/answer/…

I can't claim Autos anymore for it is expired. Is there another way to swap my autosp to autos? Could you please help?

Gyus i claimed autosp token from fjord just today so I cannot claim real tokens because i did not have it on snap period in my wallet ?? Plwase help

Hi there autos team! Although the launch gos a bit messy I’m confident that the team and community can fix it! I’ve only claimed my autosp tokens from fjord about an hour ago, so on sablier my wallet is flagged as ‘not eligible’ … Please help me out! Looking forward to reply!

I tried to claim, it said failed. Now when I try again it says Claim already been made, yet I still have $AUTOSP tokens instead of $AUTOS

Why make people go through all these fucking stress? Burn gas on fjord, burn gas to claim autosP and burn gas to finally claim That's some bullsiihit

Non voglio sapere Magari quel tubo nn andava là ( vista l'accortezza generale lo si doveva collegare ad un mezzo con su scritto autosp

From #CTAN: Bob Tennent submitted an update to the autosp package. Version: 2023-10-07 License: gpl2 Summary description: A Preprocessor that generates note-spacing commands for MusiXTeX scores ctan.org/pkg/autosp #TeXLaTeX 🎼

Incendiu la un centru de colectare si reciclare deseuri in Glina. Ard caroserii de autoturisme si material lemnos. Actioneaza 9 autosp de stingere. Sunt degajari mari de fum.

(Hilo dedicado a @oostituu -para que lo leas, z****- 🤣🤣🤣; y a @mercheiglesiasc , por su inestimable colaboración 😘😘😘😘) -Majestad, lo de Merche no es autosp...? -QUE TE CALLES!!!!