mxbai-rerank-large-v2

Name: mxbai-rerank-large-v2
Author: mixedbread-ai

by mixedbread-ai

RL-trained text reranker with state-of-the-art accuracy

475Kdl/month

560Mparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

mixedbread-ai/mxbai-rerank-large-v2

Feature URI

mixpeek://reranker@v1/mxbai_rerank_large_v2

Overview

mxbai-rerank-large-v2 is a cross-encoder reranking model from mixedbread.ai, trained using reinforcement learning from human feedback (RLHF) on top of a large transformer backbone. It achieves state-of-the-art reranking accuracy on BEIR and other standard benchmarks, outperforming many models several times its size. The model uses a cross-attention architecture that jointly encodes query-document pairs for fine-grained relevance scoring.

Architecture

Cross-encoder transformer with RLHF fine-tuning. Processes concatenated query-document pairs through full cross-attention layers, producing a single relevance score. The RL training phase uses preference data to calibrate scores toward human judgment of relevance, improving ranking quality beyond supervised-only approaches.

Mixpeek SDK Integration

// Reranking is a retriever STAGE in Mixpeek, not an ingest-time extractor.
// The rerank stage runs a cross-encoder inference service; the shipped default
// is BAAI/bge-reranker-v2-m3. Pointing it at mxbai-rerank-large-v2 means registering that
// model as a custom reranker plugin and naming it in feature_uri, which is an
// Enterprise path. Stage contract read from GET /v1/discovery/stages.
const retriever = await mx.retrievers.create({
  namespace_id: "my-namespace",
  retriever_name: "search-then-rerank",
  stages: [
    {
      stage_name: "candidates",
      stage_id: "feature_search",
      parameters: { limit: 100 },
    },
    {
      stage_name: "rerank_results",
      stage_id: "rerank",
      parameters: {
        inference_name: "BAAI__bge_reranker_v2_m3",
        query: "{{INPUT.query}}",
        document_field: "content",
        top_k: 10,
      },
    },
  ],
});