cross-encoder/ms-marco-MiniLM-L12-v2
🧠 AI Modelcross-encoder
MiniLM-L12 cross-encoder for passage re-ranking, trained on MS MARCO, fast and accurate.
The cross-encoder/ms-marco-MiniLM-L12-v2 is a state-of-the-art text ranking model that performs passage re-ranking with high accuracy. Based on the Microsoft MiniLM architecture (12 layers), it is optimized for speed and memory efficiency. The model was fine-tuned on the MS MARCO passage ranking dataset, which contains over 8.8 million passages. It outputs a similarity score between a query and a passage, enabling it to be used in two-stage retrieval pipelines. The model supports multiple frameworks including PyTorch, JAX, ONNX, and OpenVINO, and is available in safetensors format. Key features include: 384 token limit, cross-encoder architecture (joint encoding), and compatibility with sentence-transformers library for easy integration. It has been downloaded over 1.9 million times, indicating wide adoption.
💡Highlights
- ├─MiniLM-L12 architecture
- ├─Trained on MS MARCO
- └─1.9M+ downloads
🎯For
- ├─Information retrieval researchers
- ├─Search engine developers
- └─NLP practitioners