1 paper · 1 filter
Athos Georgiou
Visual document understanding typically requires separate retrieval and generation models, doubling memory and system complexity. We present Hydra, a dual-head approach that provid…