Veda Sparse ships 10% keep-ratio block-sparse attention accelerator for MiniMax H3 T2VA
linoy_tsaban · x · 2026-09-27
The Veda Sparse team released a preview checkpoint on Hugging Face that accelerates MiniMax H3 text-to-video generation:
- A 275M-param fp8 tile-score predictor selects, per layer and per head, which 128-token key tiles each query tile attends to, so attention runs block-sparse at a 10% keep ratio instead of dense
- Targets 8 denoising steps (8 NFE); the preview checkpoint was trained 600 updates on 5.17s clips only, but generalizes to 10.1s and 14.4s geometries
- fp8 storage halves download/load while top-k selection keeps accuracy unaffected; weights and tile plans ship in one file to prevent silent mispairing
- Includes deployment guides for Ada (SM89), Hopper (SM90) and Blackwell (SM120)
More from Multimodal
- Seedance 2.5 Flash Generates Intense Fight Scene From Just 2 Random Images — aziz4ai · 2026-11-16
- Runway CEO Predicts Fully AI-Generated Cultural Phenomenon by 2027 — c_valenzuelab · 2026-11-16
- Stunning AI-generated motion-blur portraits with a film-photography feel — azed_ai · 2026-11-16
- A Bauhaus functionalist prompt template for geometric poster-style AI art — LudovicCreator · 2026-11-16
- Vivix W1 video model: 100 ten-second ads for about $5 at $0.005 per second — Aiden_Tech_Ai · 2026-11-16
- User Shares Coral Reef Creation Labeled 'Qwen 4 Max (xhigh)', an Unannounced Model Name — intellectronica · 2026-11-16