Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers
Better embeddings help AI systems understand meaning more accurately, improving search, recommendation, and classification tasks across applications.
Get smart on it
Sentence Transformers version 6.0 introduced a new model type called MultiVectorEncoder for multi-vector embedding models, which match queries and documents token by token rather than compressing text into single vectors. Multi-vector models preserve fine-grained matching signals and respond well to domain-specific finetuning, particularly when documents are long, since general-purpose retrieval models often truncate documents at 180 to 512 tokens and discard most content. Finetuning involves several components: the model itself, datasets, loss functions, training arguments, evaluators, and a trainer class that can be customized for specific domains like medical, legal, or financial documents. The training approach can both finetune existing multi-vector models and build new ones from scratch on a single consumer GPU in a matter of hours.
Related news
IBM's new Granite 4.2 models ride the wave of interest in local LLMs
IBM has rolled out the newest models in its family of open-weight large language models designed to be downloaded and self-hosted. The newly launched Granite 4.2 comes in 3B, 8B, and 30B parameter variants. Like previous versions, IBM is taking a decoder-only approach here. These new releases offer a 128,000-token context window natively. The 8B and 30B variants (not the 3B one) also go through an agentic reinforcement learning block; they were trained for expanded capabilities like using the t
IBM Releases Granite 4.2: Bringing Native Reasoning and Agentic RL to Open Enterprise Models
IBM has released Granite 4.2, a family of open reasoning language models in 3B, 8B, and 30B parameter sizes. Unlike earlier Granite releases, which were instruction-following assistants, Granite 4.2 is built around explicit reasoning. Every model can emit a chain of thought before answering, and every model exposes a thinking / non-thinking switch plus a low-effort mode that spends a short reasoning budget on easy questions. The models are decoder-only dense transformers, pre-trained from scrat
