datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
ViMD_Dataset
Multi-Dialect Vietnamese: Task, Dataset, Baseline Models and Challenges (Main EMNLP 2024)
Introduction
This document presents the accompanying dataset for the paper titled "Multi-Dialect Vietnamese: Task, Dataset, Baseline Models, and Challenges". The dataset, referred to as the Vietnamese Multi-Dialect (ViMD) dataset, is a comprehensive resource designed to capture the linguistic diversity represented by 63 provincial dialects spoken across Vietnam. The paper is… See the full description on the dataset page: https://huggingface.co/datasets/nguyendv02/ViMD_Dataset.GeoMeld
🌍 GeoMeld Multi-Modal Earth Observation Dataset (WebDataset)
GeoMeld is a large-scale multi-modal remote sensing dataset introduced in our CVPRW 2026 paper on semantically grounded foundation modeling.
GeoMeld contains approximately 2.5 million spatially aligned samples spanning heterogeneous sensing modalities and spatial resolutions, paired with semantically grounded captions generated through an agentic pipeline.
The dataset is designed to support multimodal representation… See the full description on the dataset page: https://huggingface.co/datasets/vimageiitb/GeoMeld.vimeo1080pvimeo-90k-mini
Vimeo-90k-Mini
A 25% random subset of the official Vimeo-90k Triplet dataset,
used for video frame interpolation tasks.
Splits
train: ~13,000 triplets
test: ~1000 triplets
Structure
Each example contains three consecutive video frames (im1, im2, im3).
The task is typically to predict im2 given im1 and im3.
Original Dataset
Paper: Video Enhancement with Task-Oriented Flow
Authors: Tianfan Xue et al.
vimeo-90k-medium
Vimeo-90k-Medium
A 50% random subset of the official Vimeo-90k Triplet dataset,
used for video frame interpolation tasks.
Splits
train: ~26,000 triplets
test: ~2000 triplets
Structure
Each example contains three consecutive video frames (im1, im2, im3).
The task is typically to predict im2 given im1 and im3.
Original Dataset
Paper: Video Enhancement with Task-Oriented Flow
Authors: Tianfan Xue et al.
ViMedCSS-Cop
🩺 ViMedCSS: A Vietnamese Medical Code-Switching Speech Dataset (LREC 2026)
📖 Overview
ViMedCSS is a Vietnamese medical speech dataset for code-switching ASR, where each utterance contains at least one non-Vietnamese (mainly English) medical term embedded in Vietnamese speech.
📊 Dataset Statistics
Split Statistics (from ViMedCSS-Metadata)
Split
# Rows
Duration (hours)
Avg duration (s)
Total CS terms
train
11,832
24.30
7.39
12,314… See the full description on the dataset page: https://huggingface.co/datasets/shannonnonshan/ViMedCSS-Cop.ViMUL-Bench
ViMUL-Bench: A Culturally-diverse Multilingual Multimodal Video Benchmark
Overview
The evaluation toolkit to be used is lmms-eval. This toolkit facilitates the evaluation of models across multiple tasks and languages.
Key Features
🌍 14 Languages: English, Chinese, Spanish, French, German, Hindi, Arabic, Russian, Bengali, Urdu, Sinhala, Tamil, Swedish, Japanese🎭 15 Categories: Including 8 culturally diverse categories (lifestyles, festivals, foods… See the full description on the dataset page: https://huggingface.co/datasets/MBZUAI/ViMUL-Bench.vimqa
VIMQA
VIMQA is a Vietnamese dataset for advanced reasoning and explainable multi-hop
question answering. Each question requires combining facts from two different
Vietnamese Wikipedia articles, and every example ships with sentence-level
supporting facts so a model's reasoning chain can be evaluated, not just its
final answer.
The schema follows the HotpotQA
convention, so tooling written for HotpotQA transfers with minimal changes.
Usage
from datasets import… See the full description on the dataset page: https://huggingface.co/datasets/nguyenlab/vimqa.ViMD_chunked_10s
ViMD Chunked 10s — 16kHz
Preprocessed from ViMD (Nguyen et al., EMNLP 2024).
Preprocessing
Resample: 44.1kHz -> 16kHz mono
Chunking: each audio is split into consecutive NON-OVERLAPPING segments
of at most 10 seconds. ALL segments are kept, including the final
remainder (no minimum length filter). 1 original file -> ceil(len/10s) samples.
Splits: original ViMD train/valid/test kept (speaker-exclusive).
Segments of the same file always stay in the same split.… See the full description on the dataset page: https://huggingface.co/datasets/tannhoo06/ViMD_chunked_10s.ViMedAQAvi-mmarcoViMD_Dataset_Merged2026VimSketchViMDoc
ViMDoc: Visually-rich Long Multi-Document Retrieval Benchmark
ViMDoc is a benchmark for evaluating visual document retrieval in both multi-document and long document settings, proposed in our paper Hybrid-Vector Retrieval for Visually Rich Documents: Combining Single-Vector Efficiency and Multi-Vector Accuracy (ACL 2026 Findings).
🚀 Quick Start
(1) Load Queries
fromdatasets import load_dataset
dataset = load_dataset("kaistdata/ViMDoc", split="ViMDoc")
(2) Download… See the full description on the dataset page: https://huggingface.co/datasets/kaistdata/ViMDoc.vimeo-90k-test-fulltrace-forge-kimi-k3-dry-v0
trace-forge-kimi-k3-dry-v0
Reasoning traces from kimi-k3 (Moonshot native API) over a
16-prompt self-authored bank, 2 samples per prompt,
generated on 2026-07-25. All numbers in this card are measured.
Author and maintainer: Vimal Nakrani (vimalnakrani), sole author and maintainer.
Configs
The raw config has all 32 records: prompt, final answer in
content, the model's reasoning in its own reasoning field,
finish_reason, verification status, token usage, seed, and… See the full description on the dataset page: https://huggingface.co/datasets/vimalnakrani/trace-forge-kimi-k3-dry-v0.vimednliEn-VimedViMDViMLQAVietnamese portion of MLQA for monolingual QA
The train split is the machine translated train data from SQuAD v1.1.
The validation split and the test split are from MLQA.
We created this particular version with the following code:
import datasets
# Machine translated train data from SQuAD v1.1.
mlqa_train_dev = datasets.load_dataset("facebook/mlqa", "mlqa-translate-train.vi")
# Original validation and test data from MLQA
mlqa_val_test = datasets.load_dataset("facebook/mlqa", "mlqa.vi.vi")
#… See the full description on the dataset page: https://huggingface.co/datasets/phucdev/ViMLQA.vimeo_fm85vi_mc4_biology_wseg
Dataset Card for "vi_mc4_biology_wseg"
More Information needed
ViMedNLIThe dataset source is at https://aclanthology.org/2023.eacl-main.228/
I just uploaded the dataset to HuggingFace to make it easy to use.
If you use the dataset, please cite the original publication.
context_instruct_vimhopfcVIMs
Dataset Card for "VIMs"
More Information needed
vi-MedEV
Dataset Card for "vi-MedEV"
More Information needed
ViMD_north_wav2vec2ViMath_R1amazon-pricer-dataViMedAQA-database