Video DeltaNet hybrid attention speeds livestream video generation 14.5x on 8x B200
Haocheng Xi · hf · 2026-09-18
Video DeltaNet (VDN) combines local softmax attention with a bidirectional linear memory branch for long-range video context, introducing Video Delta Attention (VDA) that updates memory once per frame with spatial tokens. Staged teacher alignment injects the pathway into pretrained models; instantiated on MiniMax H3, video-to-video interactions use the hybrid while text/audio keep softmax. With 8-step distillation and an optimized SGLang serving stack, VDN-H3 denoises a 14.3-second 768p video in 6.70 seconds on eight NVIDIA B200 GPUs—a 14.5x speedup over the 50-step dense H3 baseline.
More from Multimodal
- Free course covers the 4 hard problems of production generative media agents — Al_Grigor · 2026-11-04
- PixelBench: testing multimodal models pixel-by-pixel instead of with QA — bdsqlsz · 2026-11-04
- Kitten-TTS 2 offers free voice cloning on laptops, no GPU needed — solyarisoftware · 2026-11-04
- "We'll Take Your RAM": a fully local-hardware cyberpunk meme from 2030 San Francisco — Wonderful-Yak-3348 · 2026-11-04
- Openfilm: open-source AI video agent turns coding agents into video makers — solyarisoftware · 2026-11-04
- Microsoft reopens VibeVoice: 10-second voice cloning, 90-minute local TTS — solyarisoftware · 2026-11-04