LLaVE: Large Language and Vision Embedding Models with Hardness-Weighted Contrastive Learning - View it on GitHub