This blog details SANA-Video transition to a Two-Stage Inference Paradigm. By leveraging a high-compression base model for structural discovery (Stage 1) and a step-distilled refiner for high-frequency detail injection (Stage 2), we achieve 2K resolution output without compromising the latency profile of the original 720p generation.

By reducing complexity from O(N^2) to O(N), Linear Attention is the key to processing ultra-long sequences. This post explores its mathematical core—"state accumulation"—and how it unlocks infinite context for LLMs and long video generation.

A sparse attention mechanism with O(nlogn) computational complexity for long video generation. It can speed up both training and inference by 2–4×. The code is available at https://github.com/mit-han-lab/radial-attention

SVDQuant supports NVFP4 on NVIDIA Blackwell GPUs with 3× speedup over BF16 and better image quality than INT4. Try our interactive demo below or at https://svdquant.mit.edu/! Our code is all available at https://github.com/mit-han-lab/nunchaku.

With the arrival of the RTX 5090, we built a high-performance workstation to maximize its AI computing potential. In this blog post, we share our experience—from overcoming setup challenges to testing its performance.