Sentence Transformers v6.0 Adds MultiVectorEncoder Support
The release treats ColBERT-style late interaction models as a first-class model type for training, inference, and interpretation.
Tomaarsen announced the release of Sentence Transformers v6.0. The update adds MultiVectorEncoder as a native model type that supports ColBERT-style late interaction models. These models now sit alongside dense, sparse, and reranker types for training, inference, and interpretation. Hugging Face retweeted the post. Several researchers quoted the thread, including staff from Weaviate and LightOn, noting that the change makes it simpler to train multi-vector models and that projects such as PyLate now build directly on the library.
🚨I've just released Sentence Transformers v6.0! MultiVectorEncoder joins the family: ColBERT-style late interaction models are now a first-class model type, for training, inference & interpretation, alongside dense, sparse & reranker models. Big thread 🧵
Reactions from ranked influencers
15 postsBig news!
PyLate started as a fork of Sentence Transformers to empower the community with late interaction It grew beyond expectations and became somewhat of a standard, alongside our whole ecosystem It is only logical (and for the best) that it is now merged back into ST!…
🚨I've just released Sentence Transformers v6.0! MultiVectorEncoder joins the family: ColBERT-style late interaction models are now a first-class model type, for training, inference & interpretation, alongside dense, sparse & reranker models. Big thread 🧵
Sentence Transformers is the best option to train a multi-vector model from now on ☺️ We choose to build PyLate on top of ST to benefit from all the hard work from @tomaarsen Since our work has been very convincing and impactful, it ended up back in ST, which I'm very proud of
🚨I've just released Sentence Transformers v6.0! MultiVectorEncoder joins the family: ColBERT-style late interaction models are now a first-class model type, for training, inference & interpretation, alongside dense, sparse & reranker models. Big thread 🧵
Using the best embedding/retriever models just became easier!!
🚨I've just released Sentence Transformers v6.0! MultiVectorEncoder joins the family: ColBERT-style late interaction models are now a first-class model type, for training, inference & interpretation, alongside dense, sparse & reranker models. Big thread 🧵
And STv6 supports ColPali-like models out-of-the-box, with beloved features like token pooling and similarity maps for interpretability!
Every checkpoint format loads through the same class: PyLate, Stanford-NLP ColBERT (via the HF_ColBERT marker + artifact.metadata), ColPali-style VLMs, or a bare backbone with a fresh projection. Prefixes, query expansion & the punctuation skiplist come from the saved…
Using the best embedding/retriever models just became easier!!
🚨I've just released Sentence Transformers v6.0! MultiVectorEncoder joins the family: ColBERT-style late interaction models are now a first-class model type, for training, inference & interpretation, alongside dense, sparse & reranker models. Big thread 🧵
