Hugging Face's Ettin Reranker: 6 models for accurate search reranking
Hugging Face introduced Ettin Reranker, a family of six rerankers ranging from 17 million to 1 billion parameters. The models are built on ModernBERT and traine
◐ Listen to article
Hugging Face introduced Ettin Reranker, a family of six rerankers ranging from 17 million to 1 billion parameters. The models are built on ModernBERT and trained via distillation on 143 million examples. They use Flash Attention 2 for acceleration. The smallest version runs 19 times faster than a larger competitor while losing only 0.5% in accuracy.