Sentence Transformers Adds Training Support for Multimodal Embedding and Reranker Models

Summary

Hugging Face published a blog post by Tom Aarsen detailing how to train and fine-tune multimodal embedding and reranker models using the Sentence Transformers library. This extends the library's capabilities beyond text-only embeddings to support image-text and other multimodal inputs, enabling developers to build custom retrieval and ranking models that work across modalities. Multimodal embedding models are a key component in RAG pipelines and enterprise search systems, where documents may co