Datasets:
dataset_info:
features:
- name: images
list:
image:
decode: false
- name: image_group_id
dtype: string
- name: qa
list:
- name: inputs
dtype: string
- name: choices_list
list: string
- name: labels
dtype: int64
- name: metadata
dtype: string
splits:
- name: train
num_bytes: 3782962890
num_examples: 94135
- name: validation
num_bytes: 3792161
num_examples: 100
download_size: 3786274082
dataset_size: 3786755051
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
language:
- en
license:
- mit
- cc-by-4.0
source_datasets:
- maveryn/sphinx
task_categories:
- multiple-choice
- visual-question-answering
tags:
- decision-model
- procedural
sphinx-mcq
Fresh SPHINX visual MC: charts, transformations and geometric counting.
Uses four reviewed task families, native same-scene choices and nearby count alternatives. Original panel labels stay fixed in images. Gold comes from the native constructive generator and its uniqueness guards, not a model judge. Code is MIT; shipped icon assets carry CC BY 4.0 attribution.
Original data: maveryn/sphinx, original distribution. Repackaged as parquet for tasksource by scripts/repackage_dataset/procedural/suites.py.
Conversion provenance: source revisions, hashes, configuration and exclusions.
Upstream asset credits and licenses accompany the original rendered images.
Generation checkpoints record each chunk’s seeds and exclusions. Reproduce with the resumable scale script.