---
dataset_info:
  features:
  - name: task
    dtype: string
  - name: question_id
    dtype: string
  - name: image
    dtype: image
  - name: question
    dtype: string
  - name: answer
    dtype: string
  splits:
  - name: val
    num_bytes: 1550143429.949
    num_examples: 1283
  download_size: 1284484538
  dataset_size: 1550143429.949
configs:
- config_name: default
  data_files:
  - split: val
    path: data/val-*
---

Paper: https://arxiv.org/pdf/2508.02951