AI Tools.

Search
text ranking by cross-encoder

ms-marco-MiniLM-L6-v2

Cross-encoder reranker trained on the MS MARCO passage retrieval dataset, designed to score query-document pairs jointly rather than encoding them independently. Distilled from a 12-layer cross-encoder into 6 layers to reduce latency while retaining re-ranking accuracy. Used as a second-stage ranker on top of fast first-stage retrieval (BM25 or bi-encoder).

Summary text generated by an automated pipeline from the model card · Not individually reviewed or run by us · How this page is made

Share: X Reddit HN LinkedIn

From the model card

Fields below are copied from the tags and counters on the HuggingFace repository cross-encoder/ms-marco-MiniLM-L6-v2 at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.

Publisher (HF namespace)
cross-encoder
Pipeline tag
text-ranking
Library
Sentence Transformers, Transformers
Framework tags
PyTorch, JAX
Weight formats
ONNX, safetensors, OpenVINO
License tag
apache-2.0 — read the license file in the repo before relying on it
Lineage
Language tags
English (en)
Datasets declared
sentence-transformers/msmarco
Downloads (HF counter at last fetch)
83,934,980
Likes (HF counter at last fetch)
311
Model card
https://huggingface.co/cross-encoder/ms-marco-MiniLM-L6-v2

Use cases

  • Re-ranking top-k BM25 or bi-encoder retrieval results for higher precision
  • Passage relevance scoring in RAG pipeline evaluation
  • FAQ answer ranking where accuracy outweighs latency
  • Document scoring over small pre-filtered candidate sets
  • Relevance labeling for search quality assessment

Pros

  • Joint query-document encoding yields more accurate relevance scores than bi-encoders
  • MiniLM-L6 distillation reduces inference cost vs. full 12-layer cross-encoder
  • Trained on industrial-scale MS MARCO data with established baselines
  • ONNX-compatible; Apache 2.0 license

Cons

  • Cannot index documents — must score each query-candidate pair at inference time
  • Latency scales linearly with candidate set size, impractical for large first-stage pools
  • English-only; limited accuracy on out-of-domain corpora without fine-tuning
  • Not suitable as a first-stage retriever
  • No multilingual variant at this model ID

Tags

sentence-transformerspytorchjaxonnxsafetensorsopenvinoberttext-classificationtransformerstext-rankingendataset:sentence-transformers/msmarcobase_model:cross-encoder/ms-marco-MiniLM-L12-v2base_model:quantized:cross-encoder/ms-marco-MiniLM-L12-v2license:apache-2.0text-embeddings-inferenceendpoints_compatibleregion:us