Ettin Reranker V1: Broad-Domain Text-Pair Training Data for Cross-Encoders

A training dataset used to produce the cross-encoder/ettin-reranker model family. The data is a mix of broad-domain text pairs and retrieval pairs rescored with a strong teacher reranker. All labels were produced by an automated scoring system rather than human annotators.

Use Cases

• Training cross-encoder reranker models based on the described text-pair data.
• Fine-tuning retrieval models using automatically scored query-document pairs.
• Benchmarking reranker performance on broad-domain text data.
• Studying the effectiveness of teacher-student training for reranking tasks.

Strengths

• Labels were produced by an automated scoring system, ensuring consistency.
• Data is a mix of broad-domain text pairs and teacher-rescored retrieval pairs, suggesting diversity.
• Used to train a family of models with parameters ranging from 17 million to 1 billion.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license are unknown, which may limit suitability assessment.

Provenance

Source

cross-encoder on Hugging Face

Collection Method

Mix of broad-domain text-pair data and retrieval pairs rescored with a strong teacher reranker.

Freshness

Last updated 2026-05-19 15:03:50; freshness should be verified.