Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers

Hugging Face published Tom Aarsen's guide on finetuning multimodal embedding and reranker models using Sentence Transformers. The walkthrough demonstrates finetuning Qwen3-VL-Embedding-2B for Visual Document Retrieval, improving NDCG@10 from 0.888 to 0.947 and outperforming larger models. The training pipeline reuses the same SentenceTransformerTrainer as text-only training, with datasets adding images alongside text and the model processor handling preprocessing automatically.