Inverse-LLaVA: Eliminating Alignment Pre-training Through Text-to-Vision Mapping

Open in new window