🚀 SANA-Video: Linear Attention + Constant-Memory KV Cache = Fast Long Videos 💥 Key Features 🌟 🧠 Linear DiT everywhere → O(N) complexity on video-scale tokens 🧰 Constant-memory Block KV cache → store cumulative states only (no growing KV) 🔄 🎯 Temporal Mix-FFN + 3D RoPE

Changing the autoencoder in latent diffusion models is easier than you think. 🚀 Introducing DC-Gen – a post-training acceleration framework that works with any pre-trained diffusion model, boosting efficiency by transferring it into a deeply compressed latent space with

🥳We’re releasing StreamDiffusionV2 for the live-stream community—from individual creators with one GPU to enterprise platforms with many. StreamDiffusionV2 is our follow-up to StreamDiffusion: #StreamDiffusion powered real products, but temporal consistency still bugged us.

AI is going to replace researchers? 🙀 AI is going to replace PhD students? 🙀 AI is going to take all of our jobs? 🙀 NO! But… 🚨AI is upending systems research 🚨 We show that by leveraging AI-driven research systems (ADRS), we can drastically accelerate the algorithm

What if your LLM inference automatically got faster the more you used it? Introducing ATLAS from the Together AI Turbo research team. Read more: togetherai.link/atlas Here’s Together AI Founder and Chief Scientist Tri Dao introducing ATLAS:

LLMs solving math benchmarks with verifiable answers like AIME? ✅ LLMs solving math proofs? ❌ Still an open problem. RL works great for final-answer problems, but proofs are different: - Often no single checkable answer - Correct answers can hide flawed reasoning The key

🚀 End the GPU Cost Crisis Today!!! Headache with LLMs lock a whole GPU but leave capacity idle? Frustrated by your cluster's low utilization? We launch kvcached, the first library for elastic GPU sharing across LLMs. 🔗 github.com/ovg-project/kv… 🧵👇 Why it matters:

🚀 We just set a new SOTA for LLM inference acceleration with speculative decoding. By corralling a band of specialist drafters, we got 4.99× on Llama-3.1-8B-Instruct, 4.93× on Qwen-32B — beating EAGLE3 by nearly 2x. No gimmicks. Just careful math + solid engineering. 🧵1/

1/n 🚀 Introducing SkyRL-Agent, a framework for efficient RL agent training. ⚡ 1.55× faster async rollout dispatch 🛠 Lightweight tool + task integration 🔄 Backend-agnostic (SkyRL-train / VeRL / Tinker) 🏆 Used to train SA-SWE-32B, improving Qwen3-32B from 24.4% → 39.4%

🎉 Come check out our Spotlight Poster @Neurips 2025! 🚀 Sparse VideoGen2: Accelerate Video Generation with Sparse Attention via Semantic-Aware Permutation 📍 Exhibit Hall C,D,E — #3508 🗓️ Fri, Dec 5 | 🕓 4:30–7:30 PM PST ⚡ Sparse VideoGen2 boosts video generation efficiency

See you at our poster session at NeurIPS Conference! 🎉 Radial Attention: O(n log n) Sparse Attention with Energy Decay for Long Video Generation 4:30 - 7:30pm, Exhibit Hall, CDE, id = 5414 Come talk with us if you are interested in efficient ML & VideoGen 🥳

Thrilled to release our new paper MAP: Measuring Agents in Production ⚙️🚀 2025 is the year of agents… but do they actually work in the real world? Is it just hype? A group of 25 researchers from Berkeley, Stanford, UIUC, IBM, and Intesa Sanpaolo investigated what makes agents

We've been running RadixArk for a few months, started by many core developers in SGLang LMSYS Org and its extended ecosystem (slime slime , AReaL Yi Wu). I left xAI in August — a place where I built deep emotions and countless beautiful memories. It was the best

TurboDiffusion: 100–205× faster video generation on a single RTX 5090 🚀 Only takes 1.8s to generate a high-quality 5-second video. The key to both high speed and high quality? 😍SageAttention + Sparse-Linear Attention (SLA) + rCM Github: github.com/thu-ml/TurboDi… Technical

Pass/fail benchmarks are saturated. It’s time for FrontierCS. 🚀 150+ unsolved, verifiable problems ranging from competitive programming to real-world research. Designed by PhDs & ICPC experts to evolve model intelligence. 🎓🧠 🧵👇Check it out! Paper: arxiv.org/abs/2512.15699

🚀SonicMoE🚀: a blazingly-fast MoE implementation optimized for NVIDIA Hopper GPUs. SonicMoE reduces activation memory by 45% and is 1.86x faster on H100 than previous SOTA😃 Paper: arxiv.org/abs/2512.14080 Work with Mayank Mishra, Xinle Cheng, Ion Stoica, Tri Dao