Hugging Face's logo

Datasets:
Alignment-Lab-AI
/
ARC-stuff

Dataset card Data Studio Files
xet
Community
ARC-stuff / datasets and human data /dataset /synth_riddles /readme.md
Alignment-Lab-AI's picture
Alignment-Lab-AI
Upload folder using huggingface_hub
d5bfab8 verified
preview code
|
raw
history blame contribute delete
1.28 kB

synth_riddles

This is a tiny extract of the dataset. I have cherry picked a few files from each compressed file, avoiding tasks that are too hard or broken or redundant. The compressed files are included in this repo, so you can unzip if needed.

The file: e_009d5c81_10x10_250k.tar.gz contains 250.000 files. It's one single type of task. I have encountered a few of the tasks that are invalid.

The file: rigid_15x15_depth_1_10k.tar.gz contains 10.000 files. Flip and rotate operations.

The files: rigid_and_half_depth_1_10k.tar.gz, rigid_and_half_depth_2_10k.tar.gz each contain 10.000 files. Crop in half operations.

The files: all_depth_2_10k.tar.gz, all_depth_3_10k.tar.gz, all_depth_5_10k.tar.gz, unary_depth_2_10k.tar.gz each contain 10.000 files. IMO The majority of tasks are garbage. Too complex for me to verify if they make sense or not. Most of them seems to make no sense. In ARC the biggest canvas is 30x30. However here some of the tasks have a canvas size that is bigger than 30x30, such as eee6d929.json and ed7a3761.json where the output size is 31x1. I found a few good tasks, so I have added those to this repo.