mistral-hackaton-2026/zebra-cot-mistral-small-3.2-24b-preprocessed

---
dataset_info:
features:
- name: text
dtype: string
- name: image
dtype: image
splits:
- name: train
num_bytes: 19190470516
num_examples: 174677
download_size: 15236446024
dataset_size: 19190470516
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: apache-2.0
task_categories:
- visual-question-answering
- text-generation
language:
- en
tags:
- reasoning
- chain-of-thought
- zebra-cot
- mistral
- multimodal
- hackathon
---
# Zebra-CoT Preprocessed — Mistral Hackathon 2026

Preprocessed version of the [Zebra-CoT dataset](https://huggingface.co/datasets/multimodal-reasoning-lab/Zebra-CoT) for fine-tuning Mistral-Small-3.2-24B-Instruct.

## Dataset Description
- **174,677 samples** across visual and scientific reasoning tasks
- Each sample contains an image + chain-of-thought reasoning text
- Tasks include: Chess, Maze, ARC-AGI, Ciphers, Physics, Chemistry, Graph Algorithms, Robot Planning

## Format
- `text`: formatted as `[INST] question [/INST] <think> reasoning </think> answer`
- `image`: PIL JPEG image for the corresponding visual task

## Usage
Fine-tuning Mistral-Small-3.2-24B on chain-of-thought visual reasoning.

## Hackathon
Created for **Mistral Hackaton 2026** — Fine-tuning track with W&B.