mmarco-mMiniLMv2-L12-H384-v1

The cross-encoder/mmarco-mMiniLMv2-L12-H384-v1 is a multilingual cross-encoder model for document reranking, hosted as an OpenAI-compatible API on gigarouter. It was trained on the MMARCO dataset — a machine-translated version of MS MARCO covering 14 languages — and uses the multilingual MiniLMv2 architecture as its base.

Task

Given a query and a set of candidate passages (e.g., retrieved via ElasticSearch), the model assigns a relevance score to each (query, passage) pair. Passages can then be sorted by score to produce a reranked result list.

Key Strengths

• Multilingual: trained on 14 languages (via MMARCO) and shows strong cross-lingual transfer to other languages beyond that set.
• Compact and efficient: the MiniLMv2-L12-H384 architecture balances inference speed with ranking quality.

Best For

Information retrieval pipelines that require a reranking step after an initial retrieval stage. It is suitable for applications needing multilingual or cross-lingual search, question answering, and document retrieval.

References

For implementation details and usage patterns: SBERT.net Retrieve & Re-rank.
Training code: SBERT.net Training MS Marco.