Gecko: task/query synthesis and passage relabeling for embedding distillation

What changed

Generated task descriptions and queries from web passages, retrieved alternative passages and used an LLM to relabel positives/hard negatives before contrastive embedding fine-tuning.

Scientific scope

Generated task descriptions and queries from web passages, retrieved alternative passages and used an LLM to relabel positives/hard negatives before contrastive embedding fine-tuning. The seed passage need not remain the best positive. Synthetic labels and few-shot relevance/query-likelihood judgments are not ground-truth guarantees. Teacher/data-generation and pretrained-backbone costs remain; dimension and leaderboard comparisons are source snapshots, not universal compression or current-rank claims.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. The seed passage need not remain the best positive. Synthetic labels and few-shot relevance/query-likelihood judgments are not ground-truth guarantees. Teacher/data-generation and pretrained-backbone costs remain; dimension and leaderboard comparisons are source snapshots, not universal compression or current-rank claims.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Gecko: Versatile Text Embeddings Distilled from Large Language Models

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction