Datasets:

Modalities:
Text
Formats:
parquet
License:
tabmwp / README.md
sileod's picture
Upload README.md with huggingface_hub
47ae83c verified
|
Raw History Blame Contribute Delete
1.48 kB
---
dataset_info:
features:
- name: inputs
dtype: string
- name: choices_list
list: string
- name: labels
dtype: int64
- name: metadata
dtype: string
splits:
- name: test
num_bytes: 1644805
num_examples: 1992
- name: validation
num_bytes: 1483217
num_examples: 1830
- name: train
num_bytes: 4330204
num_examples: 5317
download_size: 2809685
dataset_size: 7458226
configs:
- config_name: default
data_files:
- split: test
path: data/test-*
- split: validation
path: data/validation-*
- split: train
path: data/train-*
license: cc-by-nc-sa-4.0
source_datasets: []
tags:
- decision-model
---
# tabmwp
Native TabMWP MC questions with complete textual tables and original choices.
Original distribution: https://github.com/lupantech/PromptPG/tree/main/data/tabmwp
Dataset CC BY-NC-SA 4.0 (distinct from the code's MIT license). Native train/dev/test
partitions are preserved; tables appearing in holdouts are removed from training.
Free-answer problems supply no fabricated MC questions. Solutions remain metadata.
Original data: [original distribution](https://github.com/lupantech/PromptPG). Repackaged as parquet for [tasksource](https://github.com/sileod/tasksource) by [scripts/repackage_dataset/tabmwp.py](https://github.com/sileod/tasksource/blob/main/scripts/repackage_dataset/tabmwp.py#L41).
[Conversion provenance](provenance.json): source revisions, hashes, configuration and exclusions.