ms-marco-MiniLM-L4-v2

Model Overview

This is a cross-encoder model fine-tuned on the MS Marco Passage Ranking task for information retrieval reranking. Given a query and a set of candidate passages (e.g., retrieved via ElasticSearch), the model scores each query-passage pair and sorts passages in decreasing order of relevance.

Key Strengths

• Optimized for the reranking stage in a retrieve-and-rerank pipeline
• Compact MiniLM-L4 architecture balances speed and accuracy
• Directly outputs relevance scores for efficient ranking

Benchmark Performance

Among version 2 models, this variant achieves a strong trade-off: higher throughput than larger MiniLM-L6/L12 models while maintaining competitive ranking quality. It outperforms most version 1 and third-party models of comparable size.

Best For

• Production reranking pipelines where latency and throughput matter
• Scoring query-passage pairs for search or question answering
• Applications needing a lightweight yet effective cross-encoder

Hosted on Gigarouter

Access this model via a managed, OpenAI-compatible API — no infrastructure or model loading required.