Jev sparse attention cuts MiniMax H3 video gen time by 41.7% on an RTX 4070
CurieuxExplorer · x · 2026-09-22
Developer @sepisheim integrated the trending Jev method into MiniMax H3 video generation to sparsify attention:
- Jev scores per-layer importance across 49 layers over 4 sampling steps
- It adaptively picks sparsity rates of 1%, 3%, 5%, or 10% per layer
- On an RTX 4070, generation dropped from 6:07 to 3:34 — a 41.7% speedup
- Even with cloud queries to Jev during generation, it remains markedly faster
Related event: Jev Sparse Attention Speeds Up MiniMax H3 Video Generation by Over 40%(3 posts)→
More from Multimodal
- P2Echo: one text-prompted model segments multi-view echocardiography with high accuracy and low cost — UBC-V · 2026-11-12
- Redditor makes fast-motion short film with H3 and ships Longshot Studio v0.18.5 with face swap — R34vspec · 2026-11-12
- Reimagining Knight Rider's opening as a Netflix drama with Claude Fable 5.2 — justin_hart · 2026-11-12
- Gemini 4 Carbon builds a full 3D aquarium in Blender and Three.js in ~6 hours — sunjiao123sun_ · 2026-11-12
- Awaken Open-Sourced: Turn a Half-Duplex Speech LLM Full-Duplex With Just a Small LoRA — alexcovo_eth · 2026-11-12
- MiniMax H3 Image targets e-commerce: sketch to 4K campaign shots in ~10 seconds — MiniMax_AI · 2026-11-12