datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
mmarco-hard-negatives-reranker-filtered
mMARCO Reranker-Filtered Hard Negatives (Multilingual)
Overview
This dataset is built from mMARCO (multilingual MS MARCO) triplets for each language subset. For each (query, positive), hard negatives are bundled and then filtered using cross-encoder re-scoring. The goal is to remove negatives that are too strong or incorrect for training. The same procedure is applied to all language subsets.
The dataset is published as mmarco-hard-negatives-reranker-filtered with… See the full description on the dataset page: https://huggingface.co/datasets/hotchpotch/mmarco-hard-negatives-reranker-filtered.hpprc_emb_reranker_score
⚠️ お知らせ
よりスコア付したデータ件数とrerankerのバリエーションを増やしたデータセットのhotchpotch/hpprc_emb-scoresも公開しています。
hpprc/emb (便利なデータセットの公開、ありがとうございます)の collection と dataset がペアになっているデータに対し、negative を最大32個ランダムサンプリングしたものを、hotchpotch/japanese-bge-reranker-v2-m3-v1でスコア付けしたものです。
ライセンスは、subset ごとに hpprc/emb に記載のライセンスと同等とします。
スコア作成タイミングの revision に対してスコアを付与しているため、revision を変えると場合によって行ズレやデータ構造の変化が発生する可能性があることに注意が必要です。
例
from datasets import load_dataset
# targets = ("auto-wiki-qa", "4feb2e2492")… See the full description on the dataset page: https://huggingface.co/datasets/hotchpotch/hpprc_emb_reranker_score.reranker_continuous_filt_max7_train
Reranker training data
This data was generated using 4 steps:
We gathered queries and corresponding text data from 35 high quality datasets covering more than 95 languages.
For datasets which did not already have negative texts for queries, we mined hard negatives using the BAAI/bge-m3 embedding model.
For each query, we selected one positive and one negative text and used Qwen/Qwen2.5-32B-Instruct-GPTQ-Int4 to rate the relatedness of each query-text pair using a token "1", "2"… See the full description on the dataset page: https://huggingface.co/datasets/lightblue/reranker_continuous_filt_max7_train.reranker-scoresreranker_fulltrain_scored_filteredreranker-general-en-llm-judged
Reranker / Retriever Training Set (General-Purpose English, LLM-Judged)
Graded query-passage relevance for training rerankers (cross-encoders) and
retrievers (bi-encoders, ColBERT). English, commercial-use sources only.
Configs
pairs-graded (844k train / 8.5k test): query, passage, llm_grade (0-3), teacher_score, source, mining_method, role. For cross-encoder / reranker training.
triplets (82k train / 819 test): query, positive, negative_1..5 with teacher scores.… See the full description on the dataset page: https://huggingface.co/datasets/olaverse/reranker-general-en-llm-judged.msmarco-triplets-qwen3-rerankercqadupstack-reranker-datarerankers-and-lexical-similarities
Dataset Card for Re-ranker Evaluation Datasets
This repo contains the evaluation datasets used in the paper "Language Model Re-rankers are Fooled by Lexical Similarities" accepted to FEVER 2025.
Dataset Details
The datasets in this repo are based on the NQ, LitQA2 (from LAB-Bench) and DRUID datasets. More details on the datasets can be found in our paper.
Uses
Evaluate re-rankers.
Dataset Structure
We release the NQ, LitQA2 and DRUID… See the full description on the dataset page: https://huggingface.co/datasets/Lo/rerankers-and-lexical-similarities.Evaluation_Qwen-Qwen3-Reranker-0.6Brag-ko-bge-reranker-v2-m3-top5-answer-evalmm_reranker_rl_trainingspeaker_reranker_datamsmarco-groups-qwen3-rerankerthread-reranker-datarag-ko-jina-reranker-v2-base-multilingual-top5-answer-evalEvaluation_LoraQwen-Qwen3-Reranker-0.6Bmm_reranker_rl_trainrag-ko-bge-reranker-v2-m3-ko-top5-answer-evalenergy-eval-rag-reranker20to5