datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Nemotron-SFT-ARC-AGI-v1
Dataset Description:
Nemotron-SFT-ARC-AGI-v1 is a supervised fine-tuning (SFT) dataset of multi-turn agentic reasoning traces produced by open-weight large language models attempting to solve ARC-AGI visual-reasoning puzzles. Each ARC puzzle (a set of (input grid, output grid) demonstration pairs plus one or more test inputs, where grids are 2D integer arrays representing colors) is formatted as a text prompt and given to an agent powered by one of nine open-weight reasoning… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-SFT-ARC-AGI-v1.Nemotron-RL-ARC-AGI-v1
Dataset Description:
Nemotron-RL-ARC-AGI-v1 is a reinforcement-learning (RL) gym environment dataset of single-step ARC-AGI puzzle prompts intended for RL post-training of large language models. Each row corresponds to one ARC puzzle (a set of (input grid, output grid) demonstration pairs plus a single test input grid) rendered as a text prompt; reward is binary (1.0 / 0.0) determined by exact-match comparison against the ground-truth output grid. No LLM judge is used, no… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-ARC-AGI-v1.ARC_AGI_V1_ULTRAarc-agi-mega
ARC Mega
ARC Mega is a large-scale mixture of Abstraction and Reasoning Corpus (ARC) style prompts and non-ARC instruction/answer pairs that originated from the jack-arc training runs. The dataset is split into ARC-only and non-ARC partitions while preserving the “MindsAI” prompt/response CSV format and adding a minified JSON representation for canonical ARC tasks. Rows that contained synthetic dataset-helper prompts, system/user/GPT prefices, or malformed ARC reconstructions… See the full description on the dataset page: https://huggingface.co/datasets/mindware/arc-agi-mega.arc_agi_mini_docs
arc_agi_mini_docs — ARC-AGI v2 mini-docs ICL-QA format
Training/val/test parquets for ARC-AGI in an in-context-learning + QA
format suitable for both supervised fine-tuning and meta-learning.
Splits
Split
Rows
Source
train
13,792
ARC training (400) + ARC eval[:272] tasks, 32 augmentations each
val
99
ARC eval[272:336] (64 tasks)
test
92
ARC eval[336:400] (64 tasks)
The val/test rows include rows whose outer test query is either (a) the
raw… See the full description on the dataset page: https://huggingface.co/datasets/HerrHruby/arc_agi_mini_docs.arc-agi-2-grids
ARC-AGI-2 Grids — training + analysis corpus (NVARC-compatible)
Companion dataset for the Kaggle ARC Prize 2026 (ARC-AGI-2) solver built on
sorokin/qwen3_4b_grids15_sft139 + per-task rank-256 LoRA (NVARC lineage).
Everything here is generated from public canonical data only (1,000
training / 120 evaluation tasks); no hidden competition data is included.
Contents
Path
Rows
Description
train/train_tasks.jsonl
1,000
canonical training tasks (full I/O)… See the full description on the dataset page: https://huggingface.co/datasets/Nabidnur/arc-agi-2-grids.arc_agi_mini_docs_no_augment
arc_agi_mini_docs_no_augment — ARC-AGI v2 mini-docs ICL-QA (no augmentation)
The unaugmented counterpart to
HerrHruby/arc_agi_mini_docs.
Built from the same raw ARC-AGI files, same split assignment, same length
filter, same QA template, same leakage check — only the augmentation
expansion is disabled. Each ARC task appears as a single identity copy.
Built with:
python -m data.arc_agi.build_parquet \
--raw data/arc_agi/raw \
--out_dir <out_dir> \
--max_length 1280 \… See the full description on the dataset page: https://huggingface.co/datasets/HerrHruby/arc_agi_mini_docs_no_augment.arc-agi-augmented-100
ARC-AGI Augmented Dataset
This dataset is an augmented version of the Abstraction and Reasoning Corpus (ARC-AGI), processed for training neural networks (such as Transformers or Neural Cellular Automata).
Dataset Details
Original Source: ARC-AGI Benchmark
License: MIT
Augmentation Method:
Dihedral Transformations: 8 symmetries (rotations/flips).
Color Permutation: Random permutation of colors 1-9 (0 is fixed as background).
Translational Padding: Randomly positioning the… See the full description on the dataset page: https://huggingface.co/datasets/KotshinZ/arc-agi-augmented-100.nemotron-gym-arc-agi-transductive
laion/nemotron-gym-arc-agi-transductive
Harbor task-binary dataset (10,000 tasks) converted from nvidia/Nemotron-RL-ARC-AGI-v1 [transductive]
(part of the nvidia/Nemotron-Post-Training-v3 collection).
Each row is a valid Harbor
task binary: columns path (str) and task_binary (gzip tar). Converted with the
OpenThoughts-Agent data.nemotron_gym framework.
Grading: Exact output-grid match.