datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
MLDR
Dataset Summary
MLDR is a Multilingual Long-Document Retrieval dataset built on Wikipeida, Wudao and mC4, covering 13 typologically diverse languages. Specifically, we sample lengthy articles from Wikipedia, Wudao and mC4 datasets and randomly choose paragraphs from them. Then we use GPT-3.5 to generate questions based on these paragraphs. The generated question and the sampled article constitute a new text pair to the dataset. The prompt for GPT3.5 is “You are a curious AI… See the full description on the dataset page: https://huggingface.co/datasets/Shitao/MLDR.mldr
Dataset Card for MLDR
This is a reformatting of the MLDR dataset used to train the BGE-M3 model. See the full BGE-M3 dataset in Shitao/bge-m3-data.
Dataset Subsets
...-triplet subset
Columns: "anchor", "positive", "negative"
Column types: str, str, str
Examples:{
'anchor': '¿Cuál es el efecto de la dilución crema simple en los pelajes básicos de los caballos?',
'positive': 'Gen crema\n\nPelajes resultantes\n\nTodos los caballos tienen dos copias del gen… See the full description on the dataset page: https://huggingface.co/datasets/sentence-transformers/mldr.mldr-conteb-train
ConTEB - MLDR (training)
This dataset is part of ConTEB (Context-aware Text Embedding Benchmark), designed for evaluating contextual embedding model capabilities. It stems from the widely used MLDR dataset.
Dataset Summary
MLDR consists of long documents, associated to existing sets of question-answer pairs. To build the corpus, we start from the pre-existing collection documents, extract the text, and chunk them (using LangChain's RecursiveCharacterSplitter with a… See the full description on the dataset page: https://huggingface.co/datasets/illuin-conteb/mldr-conteb-train.Wis_Holdout_NewCZAI_Summer_School-MLDE_landscapes
CZAI Summer School 2026 — MLDE / BOES landscape data
Precomputed data backing the BOES (Bayesian Optimization in Embedding Space)
directed-evolution demo notebook for the CZAI Summer School 2026 lecture
"AI pro enzymy". See the notebook
CZAI_Summer_School-MLDE_BOES.ipynb
for how this data is used.
This repo hosts, for three combinatorial 4-site protein fitness landscapes
(GB1, PhoQ, TrpB):
*_fitness_full.csv — the full variant,fitness table, always exactly
160,000 rows — the… See the full description on the dataset page: https://huggingface.co/datasets/soldatmat/CZAI_Summer_School-MLDE_landscapes.ocr-mldr
OCR-MLDR
An OCR-degraded version of the Multi Long Document Retrieval (MLDR) benchmark, designed to evaluate embedding models on noisy, OCR-like text with long documents.
Dataset Description
A 2,000-document subsample per language was drawn from mteb/MultiLongDocRetrieval (test split).
Each passage and query was rendered as a PDF at a specific DPI / font-size setting and
re-extracted via OCR using the
ocr-robust-multilingual-embeddings
OCR simulator to introduce realistic… See the full description on the dataset page: https://huggingface.co/datasets/Psychias/ocr-mldr.mldr-conteb-eval
ConTEB - MLDR (evaluation)
This dataset is part of ConTEB (Context-aware Text Embedding Benchmark), designed for evaluating contextual embedding model capabilities. It stems from the widely used MLDR dataset.
Dataset Summary
MLDR consists of long documents, associated to existing sets of question-answer pairs. To build the corpus, we start from the pre-existing collection documents, extract the text, and chunk them (using LangChain's RecursiveCharacterSplitter with a… See the full description on the dataset page: https://huggingface.co/datasets/illuin-conteb/mldr-conteb-eval.task-specific-learningWis_ParamSweep_NewWis_Tpolymer-dynamics-Wi_0.7Wis_T_ReducedConTSG-Bench-Datasets
ConTSG-Bench Datasets (Public Subset)
This repository hosts the official public dataset subset for ConTSG-Bench.
Release Scope
The current public dataset release includes:
synth-u
synth-m
Other benchmark datasets are not publicly released at this stage.
Notes
This subset is released for open benchmarking and reproducibility.
Dataset files are organized for ConTSG benchmark pipelines.
If the HF dataset viewer is unavailable for some files, download and use… See the full description on the dataset page: https://huggingface.co/datasets/mldi-lab/ConTSG-Bench-Datasets.Wis_HoldoutFs_T_Holdoutdatasets-e2eWis_Validation_Reducedpolymer-dynamics
Dataset for polymer dynamics
References
Chen, X. et al. Constructing custom thermodynamics using deep learning. Nature Computational Science 4, 66–85 (2024).
double_pendulumDataset from 2022-07-15. There are two columns: image and trajectory. Image is the image file and trajectory is the label trajectory number_image number for the corresponding image.
so100_brickThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "so100",
"total_episodes": 1,
"total_frames": 297,
"total_tasks": 1,
"total_videos": 2,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 30,
"splits": {
"train": "0:1"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/mldev19/so100_brick.mldr-tha-qaretrieval
mldr-tha-qaretrieval
Deduplicated copy of kornwtp/mldr-tha-qaretrieval,
part of the SEA-BED data-quality work.
Source dataset: kornwtp/mldr-tha-qaretrieval
Deduplicated on: 2026-09-17
Task type: qa_retrieval
Splits: dev, test
What changed
Kept in this dataset's ORIGINAL schema -- same columns, same nesting, same extra fields (ids, titles, answers) -- so it is a drop-in replacement for the source repo. Documents differing only in case/whitespace were folded onto… See the full description on the dataset page: https://huggingface.co/datasets/puttatidam/mldr-tha-qaretrieval.so100_test_10_brick
so100_test_10_brick
This dataset was generated using a phospho starter pack.
This dataset contains a series of episodes recorded with a robot and multiple cameras. It can be directly used to train a policy using imitation learning. It's compatible with LeRobot and RLDS.
Fs_T_ReducedWis_ParamSweepUzbek_news_datasetThis is an Uzbek News Dataset with 512,750 articles (120 million words and in the Latin script) scraped from the web in 2023.
I combined and uploaded the dataset in this HF repo so that the community can fine-tune LLMs based on the Uzbek language.
@proceedings{kuriyozov_elmurod_2023_7677431,
title = {{Text classification dataset and analysis for Uzbek
language}},
year = 2023,
publisher = {Zenodo},
month = feb,
doi =… See the full description on the dataset page: https://huggingface.co/datasets/MLDataScientist/Uzbek_news_dataset.Fs_T_ParamSweepML_dataso100_test_4This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "so100",
"total_episodes": 2,
"total_frames": 596,
"total_tasks": 1,
"total_videos": 4,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 30,
"splits": {
"train": "0:2"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/mldev19/so100_test_4.Fs_Validation_ReducedMLD-VC
🎥 MLD-VC: Multimodal Dataset for Video Conferencing
When AVSR Meets Video Conferencing: Dataset, Degradation, and the Hidden Mechanism Behind Performance Collapse (CVPR 2026)
📄 [Paper] | 🤗 [Hugging Face Dataset]
📌 Overview
MLD-VC is the first multimodal dataset specifically designed for Audio-Visual Speech Recognition (AVSR) in real-world video conferencing (VC) scenarios.
Unlike traditional AVSR datasets collected in controlled offline environments, MLD-VC… See the full description on the dataset page: https://huggingface.co/datasets/nccm2p2/MLD-VC.