alexshengzhili/llava-7bv0-mm-projector-ft-with-ocr-caption-prompted-paragraph

This is the feature alignment pre-training work to train only only the multi-modal projector. "Predict" paragraph given caption, ocr and image token