GIFARC 

Synthetic Dataset for Leveraging Human‑Intuitive Analogies to Elevate AI Reasoning

     

By embedding robust human-intuitive analogies into ARC-style tasks, GIFARC guides AI agents to evaluate the task analogically before engaging in brute-force pattern search, thus efficiently reducing problem complexity and build a more concise and human-understandable solution.

---

will turn into .... GIFARC!

TL;DR

• 1,614 ARC style puzzles made from GIF with analogy.
• Pair‑wise ground‑truth mappings + rich textual rationales for supervised or in‑context use.
• Easy Play generation pipeline - extend or remix new analogy families with gif in a few minutes.
• Friendly Hugging Face dataset & interactive web demo for instant exploration.

Table of Contents

1. Quick Start
2. Dataset Card
3. Pipeline Overview
4. Project Structure
5. Citing GIFARC
6. Acknowledgements
7. License

---

Quick Start

1. Install

We highly command to using docker. To setting with docker check SETUP.md.

git clone <GIT_url>
cd gifarc
pip install -r requirements.txt
pip install -r requirements-dev.txt      

2. Pull the Dataset

from datasets import load_dataset
ds = load_dataset("DumDev/gif_arc")

3. Generate Your Own GIFARC

Once your Set up is down, open description_executor.ipynb and run the code here.

4. Check the Web Demo

GIFARC Web Demo.

---

Dataset Card

Split	#Tasks	#Unique GIFs	Size
Train	1,614	1,614	< 100 MB

Every task packages looks as follows:

{
  "source": "<source code>", # python code string
  "examples": [
      [<input_grid_1>,<output_grid_1>], # pair 1
      [<input_grid_2>,<output_grid_2>], # pair 2
      ...
    ], 
  "seeds": [
      "<file_name_1>",
      "<file_name_2>",
      ...,
      "<file_name_N>",
      "<Concept_and_description>"
    ], 
  "url": "<minified_url>"
}

See the full dataset card for licensing, intended use, and data statements.

---

Pipeline Overview

• Modular & Easy generation – After put GIF in data/GIF, just click all run button at description_executor.ipynb to generate Your own data!
• Stable environment setting enable easy set up with docker and devcontainer.
• All intermediate artifacts are cached for reproducibility.

Detailed instructions live in GENERATION.md.

---

## Project Structure

./GIFARC
├── data
│   └── GIF
├── description_executor.ipynb # use this to execute
├── docker-compose.yml
├── docs
│   ├── EXPERIMENTS.md
│   ├── GENERATION.md
│   ├── project_directory_tree.txt
│   └── SETUP.md
├── loggings
├── README.md
├── requirements-dev.txt
├── requirements.txt
├── results # this will generate automatically
└── src
    ├── execution.py
    ├── experiments.py
    ├── generate_descriptions.py
    ├── generate_problems.py
    ├── generate_visualization_html.py
    ├── GIFARC_data_batch
    ├── GIFARC_utils
    ├── misc
    ├── parse_batch_description_samples.py
    ├── prompts
    ├── seeds
    ├── utility
    └── visualize_problems.py

---

Citing GIFARC

@misc{gifarc2025,
  title   = {GIFARC: Synthetic Dataset for Leveraging Human-Intuitive Analogies to Elevate AI Reasoning},
  author  = { Anonymous },
  year    = {2025},
  note    = {Under review at NeurIPS Datasets & Benchmarks 2025},
  url     = {}
}

---

Acknowledgements

• GIPHY for powering the GIF search API.
• BARC – our generation pipeline stands on the shoulders of this excellent project.
• GIFARC wouldn’t be possible without the open‑source community and our amazing reviewers.

---

License

Distributed under the MIT License.