🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Thanks for having me on, it was really fun!

chipmunk is up on arxiv! across HunyuanVideo and Flux.1-dev, 5-25% of the intermediate activation values in attention and MLPs account for 70-90% of the change in activations across steps caching + sparsity speeds up generation by only recomputing fast changing activations

🥳 Happy to share our new work –  Kinetics: Rethinking Test-Time Scaling Laws 🤔How to effectively build a powerful reasoning agent? Existing compute-optimal scaling laws suggest 64K thinking tokens + 1.7B model > 32B model. But, It only shows half of the picture! 🚨 The O(N²)

Great work from Infini-AI-Lab! Congrats Beidi Chen!

When we put lots of text (eg a code repo) into LLM context, cost soars b/c of the KV cache’s size. What if we trained a smaller KV cache for our documents offline? Using a test-time training recipe we call self-study, we find that this can reduce cache memory on avg 39x

Excited to be presenting our new work–HMAR: Efficient Hierarchical Masked Auto-Regressive Image Generation– at #CVPR2025 this week. VAR (Visual Autoregressive Modelling) introduced a very nice way to formulate autoregressive image generation as a next-scale prediction task (from

Announcing HMAR - Efficient Hierarchical Masked Auto-Regressive Image Generation, led by Hermann! HMAR is hardware-efficient, reformulates autoregressive image generation in a way that can take advantage of tensor cores. Hermann is presenting it at CVPR this week!

1/ Model architectures have been mostly treated as fixed post-training. 🌱 Introducing Grafting: A new way to edit pretrained diffusion transformers, allowing us to customize architectural designs on a small compute budget. 🌎 grafting.stanford.edu Co-led with Michael Poli

And to close out a trio of diffusion papers… Super excited to announce Grafting - a method for distilling pretrained diffusion transformers into *new architectures*, led by Keshigeyan Chandrasegaran! Swap attention for new primitives for 2% pretraining cost, exciting for modeling research!

Scale alone is not enough for AI data. Quality and complexity are equally critical. Excited to support all of these for LLM developers with Snorkel AI Data-as-a-Service, and to share our new leaderboard! — Our decade-plus of research and work in AI data has a simple point:

Never a better time to work with Snorkel AI :)

Chipmunks can now hop across multiple GPU architectures (sm_80, sm_89, sm_90). You can get a 1.4-3x lossless speedup when generating videos on A100s, 4090s, and H100s! Chipmunks also play with more open-source models: Mochi, Wan, & others (w/ tutorials for integration) 🐿️

Chipmunks for everyone!

What a throwback to weak supervision! Great work Jon Saad-Falcon Kelly Buchanan Mayee Chen!

🐿️ chipmunk ship! flux kontext supported for up to 30% faster cute chipmunks!

Day zero support for Flux kontext dev on Chipmunk! Great work Austin Silveria!

Happy to share that our HMAR code and pre-trained models are now publicly available. Please try them out here: code: github.com/NVlabs/HMAR checkpoints: huggingface.co/nvidia/HMAR

HMAR code and models are out!