Voyage's rerankers start taking instructions at query time
- Source
- x.com
- Date
📣 Announcing rerank-2.5 and 2.5-lite: our latest generation of rerankers! • First reranker with instruction-following capabilities • rerank-2.5 and 2.5-lite are 7.94% and 7.16% more accurate than Cohere Reranker v3.5 • Additional 8.13% and 7.55% performance gain with instruction
📈 The rerank-2.5 series sets a new price-performance frontier even w/o instruction-following. rerank-2.5 improves retrieval quality over rerank-2 by 1.85%, & 2.5-lite improves over 2-lite by 3.40%, while increasing to a 32K context length & maintaining their respective token price.
⬆️ When applied as a second-stage retrieval method atop the strongest embedding model (voyage-3-large), both rerank-2.5 and rerank-2.5-lite are the only rerankers to consistently improve performance across all domains.
🎉 Both models are available today! The first 200 million tokens are free. Get all the details in our latest blog: https://t.co/XoZRrsbFL0
Context
Voyage AI's rerank-2.5 models let a search application explain what relevance means for a particular request. A reranker receives candidate documents from an initial search and changes their order. The release article describes adding plain-language instructions before or after the query so those instructions influence the relevance scores.
For example, a legal search could prioritize statutes over court cases, even when both concern the requested topic. This makes document type or application part of the ranking decision, giving developers a way to steer results without expressing their whole preference through the search terms alone.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- reranking — A second pass that re-scores retrieved candidates by reading each one against the query, fixing the ordering that fast vector search got approximately right.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
Checking sign-in…
Loading comments…