Granite Embedding Reranker English R2
Granite Embedding Reranker English R2 is a 149M ModernBERT cross-encoder. It scores a query and document together to refine an existing result list. It is not an index-building vector encoder.
Model design and output
Training uses a ranking objective including pListMLE and model merging. Unlike a bi-encoder, the reranker processes each query-document pair jointly. Its maximum input length therefore covers the query and the candidate document together.Granite Embedding Reranker English R2 model card
Practical use
Practical guidance. Apply the reranker to a manageable candidate set after retrieval. Measure the improvement in top results against the added per-query computation. Preserve the original document identifiers so reranked evidence can still be traced to the stored source.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | Downloadable |
| Model identifiers | ibm-granite/granite-embedding-reranker-english-r2 |
| Context window | 8,192 tokens Combined query-document input budget. Granite Embedding Reranker English R2 model card |
| License | Apache 2.0Granite Embedding Reranker English R2 model card |
| Output | Query-document relevance scoreGranite Embedding Reranker English R2 model card |
Related models
- Granite Embedding English R2
IBM · Granite Embedding
A 149M English retrieval model that produces 768-dimensional vectors.
Launch date Downloadable - Granite Embedding Small English R2
IBM · Granite Embedding
A 47M ModernBERT English retrieval encoder that produces 384-dimensional vectors.
Launch date Downloadable