https://chiosa.app/papers/curvetq-rotation-free-trellis-quantization-of-llm-weights-via-curvature-weighted-search-2610-09212Chiosaen2026-10-11T00:15:48ZCurvature-guided compression avoids rotating model weightshttps://chiosa.app/papers/grace-generation-aware-latent-compression-for-efficient-video-generation-2610-10524Chiosaen2026-10-10T14:41:28ZVideo compression cuts generation time while preserving Wan2.1 scoreshttps://chiosa.app/papers/inverting-multi-vector-visual-document-indices-2610-09920Chiosaen2026-10-10T14:41:28ZStored page vectors can reveal the document they representhttps://chiosa.app/papers/real-time-joint-audio-video-generation-by-parallel-adapter-composition-2610-10343Chiosaen2026-10-10T14:41:28ZTwo adapters let audio–video generation stream in a few stepshttps://chiosa.app/papers/scalable-patch-level-self-supervised-learning-2610-10013Chiosaen2026-10-10T14:41:28ZPatch-level visual learning scales with a simpler objective