license: mit
task_categories:
- image-to-text
- text-to-image
language:
- en
pretty_name: simons ARC (abstraction & reasoning corpus) solve span version 12
size_categories:
- 1K<n<10K
configs:
- config_name: default
data_files:
- split: train
path: data.jsonl
Version 1
ARC-AGI Tasks where the job is to transform images that have intersection patterns between row/column spans.
example count: 2-3.
test count: 1-2.
image size: 3-8.
span count: 3-5.
Version 2
image size: 3-10.
span count: 3-6.
Version 3
image size: 3-12.
span count: 3-7.
Version 4
Focus only on generate_task_with_template_lines.
image size: 4-8.
span count: 4-5.
Version 5
Focus only on generate_task_with_template_lines.
image size: 4-10.
span count: 4-6.
Version 6
Enabled both generate_task_with_intersecting_spans and generate_task_with_template_lines.
image size: 3-12.
span count: 3-7.
Version 7
Focus only on generate_task_with_alternate.
image size: 5-8.
span count: 5-7.
Version 8
Focus only on generate_task_with_alternate.
image size: 5-10.
span count: 5-7.
Version 9
Focus only on generate_task_with_alternate.
image size: 5-12.
span count: 5-7.
Version 10
Enabled all transformations.
Earlier predictions added to some of the rows.
Version 11
Smaller images.
Version 12
Added fields: arc_task, test_index, earlier_output.