Introducing the Ettin Reranker Family

The Ettin Reranker Family introduces six new Sentence Transformers CrossEncoder models ranging from 17M to 1B parameters, achieving state-of-the-art performance at their respective sizes.

• •Six model variants (17m, 32m, 68m, 150m, 400m, 1b) built on Ettin ModernBERT encoders supporting up to 8,192 tokens of context length
• •Trained via distillation on mixedbread-ai/mxbai-rerank-large-v2 scores for efficient knowledge transfer from larger teacher models
• •Implements retrieve-then-rerank pipeline combining fast embedding models for candidate retrieval with accurate reranking
• •Supports Flash Attention 2 and bfloat16 optimizations delivering 1.7x-8.3x speedup improvements
• •Achieves state-of-the-art MTEB (eng, v2) Retrieval benchmark results across multiple embedding model pairings