ettin-reranker-68m-v1

Name: ettin-reranker-68m-v1
Author: cross-encoder

by cross-encoder

68M-param cross-encoder reranker balancing accuracy and speed

N/Adl/month

68Mparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

cross-encoder/ettin-reranker-68m-v1

Feature URI

mixpeek://reranker@v1/cross_encoder_ettin_68m_v1

Overview

Ettin Reranker 68M is the mid-size model in the Ettin family, built on the ModernBERT encoder architecture and distilled from mxbai-rerank-large-v2. It provides a balance between the tiny 17M/32M variants and the larger 150M/400M/1B models, making it suitable for latency-sensitive production workloads that still need strong reranking quality.

Architecture

Cross-encoder architecture using ModernBERT as the backbone. Concatenates query and document into a single input sequence, processes through bidirectional attention layers, and outputs a single relevance score. Trained via knowledge distillation from larger reranking models with margin-MSE loss.

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 ettin-reranker-68m-v1 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,
      },
    },
  ],
});