Song Han

[Song Han @songhan_mit]

ForeAct: Steering Your VLA with Efficient Visual Foresight Planning

[Zhuoyang Zhang @zhuoyang_zhang]

We‘ll be presenting ForeAct (🌟Highlight) at CVPR 2026:
📍 Poster: Sunday, 3:30PM, ExHall A #95
📍 NVIDIA Tech Talk: Friday, 12:40PM, Booth #211 @NVIDIAAI
Feel free to stop by and chat! Also find our coffee making demo empowered by ForeAct! (https://t.co/a58GwCbgR0)

I’ll be at #CVPR2026 from Jun 4-7. Open to chat about VLA, world models and efficient visual generation!

[Shuo Yang @Andy_ShuoYang]

FlashLib update: we now support ANN search with IVF-Flat — up to 6.5× faster than cuVS on real-world vector workloads (SIFT-1M) while matching recall.

LEANN now supports FlashLib as a backend: 26× faster build, 29× faster single-query, and 298× faster batch search. Huge thanks to @YichuanM for the help!

We’re also opening Discord / Slack channels — join us to suggest new operators you want to see, and hardware backends you want FlashLib to support next!

Slack: https://t.co/BiH46PvPbH

Discord: https://t.co/6sfTJKkLtG

[Yukang Chen @yukangchen_]

We released a blog on "Why Video Gen Is an Infra Problem".
https://t.co/VZFb6L838c

We discuss why long video generation requires full-stack co-design across models, memory, KV cache, VAE decoding, scheduling, and deployment infrastructure, with LongLive 2.0 (https://t.co/QXF2lfolpj) as a case study.

[NVIDIA AI @NVIDIAAI]

Introducing Cosmos 3: Our latest frontier model for Physical AI

Cosmos 3 is the world’s first fully open omnimodel with native vision reasoning, world and action generation.

Today we’re releasing Super (32B) and Nano (8B) variants.

[Song Han @songhan_mit]

SANA Streaming: V2V on a single 5090

[Song Han @songhan_mit]

Hide to Guide: Learning via Semantic Masking

[rita @rita17371931]

Expert traces can guide RLVR, but full traces may leak key values, executable code, or answer entities — creating an unintended reward-hacking channel.

We propose SMEPO: Semantic Masked Expert Policy Optimization.

Paper: https://t.co/55uoo9rGTw
Code: https://t.co/AH8BiTdpPc

[Shuo Yang @Andy_ShuoYang]

Flash-KMeans was only the beginning.

Today, from the Flash-KMeans team, we are releasing FlashLib — a GPU library for fast, predictable, agent-ready classical ML operators.

Up to 26× on KMeans, 19× on KNN, 40× on HDBSCAN, 208× on TruncatedSVD, 47× on PCA, 147× on exact t-SNE, and 49× on MultinomialNB over state-of-the-art (cuML).

Blog: https://t.co/P31SGl0cyT
Code: https://t.co/9nkO2hmeOl

[Yukang Chen @yukangchen_]

🚀 LongLive 2.0 just got faster!
Since last week’s release, we further optimized the NVFP4 inference path and improved the overall throughput by 18.6%.

🔥Now, generating a 64s video takes only 30.6s end-to-end, including VAE decoding.
⚡⚡That’s over 2× real-time generation.

🛠️ What changed under the hood?
• Fused Triton RoPE / adaLN kernels
• Reduced KV-cache synchronization overhead
• In-place quantized KV-cache updates
• Faster FP4 KV dequantization
• Pinned VAE transfers
• Safer LoRA-before-quantization setup

🎬 LongLive 2.0 is our open-source 4-bit long-video generation infra for both training and inference.

🚀 We are continuing to push long-video generation toward faster, lighter, and more practical deployment.

🔗 Code: https://t.co/QXF2lfnNzL

#LongVideoGeneration #VideoGeneration #Realtime #AIInfra #EfficientAI #FP4 #Parallel #NVIDIA

[Song Han @songhan_mit]

Explore our kernel design agents:

[NVIDIA AI @NVIDIAAI]

Long video generation is a systems problem.

Introducing LongLive-2.0 from NVIDIA Research: an end-to-end NVFP4 training and inference system for long video generation.

Low-precision deployment often relies on post-training quantization, creating a gap between how models are trained and how they run.

LongLive-2.0 aligns NVFP4-aware training, distillation, and W4A4 inference, maintaining strong benchmark quality while improving speed and memory efficiency.

[Pavlo Molchanov @PavloMolchanov]

We’re releasing Nemotron-Labs-Diffusion - the first Tri-mode LM family (3B/8B/14B) that switches between 1⃣Autoregressive, 2⃣Diffusion, and 3⃣Self-Speculation decoding by simply changing the attention pattern/mask.

One model Three decoding modes. No extra draft models. No architecture changes. Just significantly better efficiency across different concurrency levels.

Up to 4× higher real throughput for a single user.

🤗 HF Collection: https://t.co/1zStcCCWPi, open license
🛜 Project page: https://t.co/y6TEAvLFvD
📰 Tech report: https://t.co/NSjKxEyHnT

Details below 👇

[Song Han @songhan_mit]

LongLive 2.0, nvfp4 infra for long video generation:

[Yukang Chen @yukangchen_]

🚀 Excited to release LongLive 2.0!
🎬 An end-to-end infrastructure for long video generation, with FP4 and parallelism at the core of both training and inference.

⚡45.7 FPS generation speed on 5B model⚡

✨ LongLive 2.0 supports real-video training, few-step distillation, multi-shot training/inference, sequence-parallel acceleration, NVFP4 KV cache, and async VAE decoding deployment.

🧩 To our knowledge, this is the first open-source 4-bit long video generation infra that covers both training and inference.

🙌 Welcome to check it out, try it, and share feedback!

🔗 Code: https://t.co/QXF2lfnNzL
📰 Paper: https://t.co/gKtarHj17c
🎥 Demo: https://t.co/RLF1wfOXVZ

#LongVideoGeneration #VideoGeneration #Realtime #AIInfra #EfficientAI #FP4 #Parallel #NVIDIA

[Song Han @songhan_mit]

SANA world model, small and fast:

[NVIDIA AI @NVIDIAAI]

One image + text + camera trajectory = controllable worlds. All on a single GPU.

Our research team just released SANA-WM, a 2.6B open source world model natively trained for 60-second video generation with precise camera control.

[Jiayi Yuan @jiayiy]

🚀 BLASST just won Best Paper at #MLSys26!
In this paper, we introduce a simple, training-free dynamic sparse attention mechanism that uses a single scalar threshold on online softmax statistics to skip negligible attention blocks.
Unfortunately I won’t be there in person, but please say hi to my awesome coauthors! 🙌
Paper: https://t.co/HDDHDqKRTT

[Huizi Mao @huizi_mao]

Glad to be featured by SemiAnalysis. Our work BLASST was also selected as MLSys 2026 Best Paper: https://t.co/OlkQ7x75BN

[Song Han @songhan_mit]

Explore SANA World Model, using hybrid linear attention, efficient and fast!

[Haoyi Zhu @HaoyiZhu]

🤩Excited to share SANA-WM: a 2.6B open-source world model for minute-scale 720p video generation.

Given one image + text + a 6-DoF camera trajectory, it synthesizes action-controllable 60s worlds on a single GPU.

Project: https://t.co/5NINfiFoTK
Paper: https://t.co/JKczmyRsJL

[Song Han @songhan_mit]

Explore AnyFlow for any step distillation, accelerating video diffusion model inference:

[YUCHAO GU @YuchaoGu]

🚀 We are excited to announce the release of AnyFlow, the first any-step video diffusion on-policy distillation (OPD) framework. By leveraging Flow Map distillation, AnyFlow significantly enhances model inference efficiency by reducing sample steps. (Code, models, and demos are now open-source!)

Key Highlights:
⚡ Any-Step Generation: Unlike traditional distilled models tied to fixed step budgets, AnyFlow enables a single model to adapt to arbitrary inference budgets. It achieves high-quality few-step generation while providing stable improvements as more sampling steps are added.
🔀 Multiple Architectures: AnyFlow supports any-step distillation for both causal and bidirectional video diffusion models.
🎬 Multiple Tasks: AnyFlow supports Text-to-Video, Image-to-Video, and Video-to-Video generation within one causal video diffusion model.
📈 Scalable Performance: AnyFlow is validated from 1.3B up to 14B parameters.

📄 Paper: https://t.co/Qqik8l29oB
💻 Code: https://t.co/KOMv9RtuWu
🎨 Pre-trained Models: https://t.co/Br1MNllUu8
🎬 Demo: https://t.co/hxbl56lPFU

[DailyPapers @HuggingPapers]

NVIDIA just released AnyFlow on Hugging Face

The first any-step video diffusion model that generates high-quality text-to-video with any inference budget - 4 steps or 50, quality scales smoothly without degradation.

[Song Han @songhan_mit]

Explore lightening OPD for efficient LLM post training:

[Han Cai @hancai_hm]

On-policy distillation (OPD) is one of the most effective LLM post-training methods, but it traditionally requires a costly live teacher server throughout training.

In our latest work, Lightning OPD, we show that OPD can be performed fully offline by precomputing teacher log-probabilities before training, reducing OPD to a standard single-model training job.

A key insight is Teacher Consistency: the SFT teacher and OPD teacher must be identical. Otherwise, offline OPD suffers a significant accuracy drop.

Paper: https://t.co/zTW3mm2wWX
Code: https://t.co/7iOB8LezVQ

Contributors: Yecheng Wu, Song Han, Han Cai

[Zhijian Liu @zhijianliu_]

DFlash for Gemma 4: Up to 6x Faster. ⚡⚡

Great to see MTP land natively in Gemma 4 today. If you want to push it further, try DFlash — open source, same quality, more speed!!

https://t.co/wKcRoibuOB