Hugging Face AI Updates: August 27, 2026

1. Sentence Transformers v6.0 Adds Multi-Vector Embedding Training

Hugging Face. A Hugging Face blog post details Sentence Transformers v6.0, which introduces MultiVectorEncoder as a fourth model type alongside its embedding and reranker options. The class trains ColBERT-style late interaction models that keep one vector per token and score queries against documents using the MaxSim operator. A featured medical model trained on 1 million question-passage pairs from the MIRIAD dataset reached 0.9139 NDCG@10 on a held-out evaluation of 1,000 questions against 200,000 passages, using 14.5 hours on a single RTX 3090 GPU. Applying 1-bit PLAID quantization reduced embedding storage to 3.37 GB while retaining 0.8984 NDCG@10. Source