license: mit
task_categories:
- image-to-text
- text-to-image
language:
- en
pretty_name: simons ARC (abstraction & reasoning corpus) solve count version 9
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 count N number of pixels in the input, and in the output repeat a pattern N times.
example count: 3-4.
test count: 1-2.
input image size: 3-8.
output pattern image size: 1-3.
pixel count: 1-3.
I had a serious mistake in number_of_positions where I didn't deal with clashing xy coordinates, causing the pixel count to not match with the pattern count in the output.
Version 2
input image size: 3-10.
pixel count: 1-4.
I had a serious mistake in number_of_positions where I didn't deal with clashing xy coordinates, causing the pixel count to not match with the pattern count in the output.
Version 3
input image size: 3-12.
output pattern image size: 1-4.
pixel count: 1-5.
I had a serious mistake in number_of_positions where I didn't deal with clashing xy coordinates, causing the pixel count to not match with the pattern count in the output.
Version 4
Fixed the number_of_positions code. So the input count and output count now should match.
input image size: 3-14.
output pattern image size: 1-5.
pixel count: 1-6.
Version 5
Focus only on generate_task_count_pixels_and_repeat_input_pattern.
input image size: 4-8.
input pattern image size: 2.
pixel count: 1-3.
Version 6
Focus only on generate_task_count_pixels_and_repeat_input_pattern.
input image size: 4-12.
input pattern image size: 2-3.
pixel count: 1-4.
Version 7
Focus only on generate_task_count_pixels_and_repeat_input_pattern.
input image size: 4-14.
input pattern image size: 2-4.
pixel count: 1-5.
Version 8
Added fields: arc_task, test_index, earlier_output.
Version 9
Replaced RLE compressed response with raw pixel response.