Japanese Reranker V2 Hard Negatives Scores: Training Data for Reranker Models

hotchpotch/japanese-reranker-v2-hard-negatives-scores contains unified hard-negative score rows used to train the Japanese Reranker v2 family. The dataset stores teacher scores as floating-point soft labels and follows the HPPRC Emb Score style, with scores and referenced document text stored separately. It was last updated on 2026-06-03.

Use Cases

• Training Japanese reranker models based on hard-negative scores.
• Fine-tuning ranking algorithms using floating-point soft labels.
• Conducting experiments on Japanese text retrieval and ranking.

Strengths

• Dataset was used to train the Japanese Reranker v2 family of models.
• Stores teacher scores as floating-point soft labels, which can provide richer training signals.

Limitations

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

Provenance

Source

hotchpotch

Collection Method

Generated from the training sources used for Japanese Reranker v2-family experiments.

Freshness

Last updated 2026-06-03 21:57:09.