inverse-llava-text-to-vision

Map text embeddings into visual representation space for multimodal fusion, eliminating expensive image-text alignment pre-training while improving reasoning-heavy tasks by up to 27.2%.

Independent GitHub snapshot · Automated classification · Original documentation governs