CoolFace
Datasetpublic

raoanmol/ViTaB-A

ViTaB-A Dataset A normalized table question answering dataset for the ViTaB-A research project. Configs hitab: Derived from HiTab (10,670 samples) fetaqa: Derived from FeTaQA (10,330 samples) Usage from datasets import load_dataset hitab = load_dataset("raoanmol/ViTaB-A", "hitab") fetaqa = load_dataset("raoanmol/ViTaB-A", "fetaqa") Schema Each sample contains: Field Type Description id string Unique identifier (e.g.… See the full description on the dataset page: https://huggingface.co/datasets/raoanmol/ViTaB-A.

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face
0likes30downloads
Dataset Card

ViTaB-A Dataset

A normalized table question answering dataset for the ViTaB-A research project.

Configs

  • —hitab: Derived from HiTab (10,670 samples)
  • —fetaqa: Derived from FeTaQA (10,330 samples)

Usage

python
from datasets import load_dataset

hitab = load_dataset("raoanmol/ViTaB-A", "hitab")
fetaqa = load_dataset("raoanmol/ViTaB-A", "fetaqa")

Schema

Each sample contains:

FieldTypeDescription
idstringUnique identifier (e.g. vitaba_000001_hitab)
splitstringDataset split (train/validation/test)
questionstringNatural language question about the table
answerlist or stringAnswer (list for HiTab, string for FeTaQA)
citationlist[str]Excel-style cell references (e.g. ["=E7"])
table_jsondictSimplified table with keys: title (string), header (list of header rows), rows (list of data rows)
table_mdstringMarkdown representation of the table with Excel-style row/column labels
table_imagesdictTable images as base64 PNGs. Keys: arial, times_new_roman, red, blue, green. Unrendered variants are empty strings.
sourcestringSource dataset and split (e.g. hitab_train)
source_idstringOriginal ID from source dataset