Machine Heart
Nov 5, 2026 · Artificial Intelligence
Multimodal Flow: Fully Continuous Unified Language-Vision Modeling
Researchers from HUST and Horizon Robotics propose Multimodal Flow, a fully continuous multimodal architecture that unifies language and vision in shared embedding spaces using Flow Matching and Hyperchunks, outperforming discrete and hybrid approaches on generation and understanding benchmarks with only 1.6B parameters and 150B tokens.
Continuous EmbeddingFlow MatchingFoundation Models
0 likes · 12 min read
