Are Your Reasoning Models Reasoning or Guessing? A Mechanistic Analysis of Hierarchical Reasoning Models
Paper • 2601.10679 • Published
pad_id int64 0 0 | ignore_label_id int64 0 0 | blank_identifier_id int64 0 0 | vocab_size int64 6 12 | seq_len int64 81 900 | num_puzzle_identifiers int64 1 1.05M | total_groups int64 960 1.12k | mean_puzzle_examples float64 1 4.15 | sets listlengths 1 1 |
|---|---|---|---|---|---|---|---|---|
0 | 0 | 0 | 12 | 900 | 1,045,832 | 1,120 | 4.151491 | [
"all"
] |
0 | 0 | 0 | 12 | 900 | 876,405 | 960 | 4.145814 | [
"all"
] |
0 | 0 | 0 | 6 | 900 | 1 | 1,000 | 1 | [
"all"
] |
0 | 0 | 0 | 11 | 81 | 1 | 1,000 | 1 | [
"all"
] |
0 | 0 | 0 | 11 | 81 | 1 | 1,000 | 1 | [
"all"
] |
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
license: apache-2.0 task_categories: - question-answering tags: - sudoku - reasoning - hierarchical-reasoning-model - puzzle-solving pretty_name: Sudoku-Extreme & Augmented Datasets for HRM Reproduction
Sudoku-Extreme Datasets for HRM Reproduction
Preprocessed datasets used for reproducing the Hierarchical Reasoning Model (HRM) and Augmented HRM papers on the Sudoku-Extreme benchmark.
Datasets
| Dataset | Description | Train Examples | Test Examples |
|---|---|---|---|
sudoku-extreme-1k-aug-1000 |
Vanilla Sudoku-Extreme (1000 puzzles, 1000x augmented) | 1,001,000 | 422,786 |
sudoku-extreme-1k-aug-1000-hint |
Augmented version with easier puzzles mixed in | 2,002,000 | 422,786 |
Source Papers
- HRM: Hierarchical Reasoning Model (Wang et al., 2025)
- Augmented HRM: Are Your Reasoning Models Reasoning or Guessing? (Ren & Liu, 2026)
Dataset Format
Each dataset contains train/ and test/ directories with:
all__inputs.npy— Puzzle inputs, shape(N, 81), values 1-10all__labels.npy— Solution labels, shape(N, 81), values 2-10all__puzzle_indices.npy— Cumulative indices marking puzzle boundariesall__puzzle_identifiers.npy— Puzzle type IDsall__group_indices.npy— Cumulative group boundary indicesdataset.json— Metadata (vocab_size=11, seq_len=81, pad_id=0)
How to Build From Scratch
# Vanilla dataset
python dataset/build_sudoku_dataset.py --output-dir data/sudoku-extreme-1k-aug-1000 --subsample-size 1000 --num-aug 1000
# Augmented (hint) dataset
python dataset/build_sudoku_dataset.py --output-dir data/sudoku-extreme-1k-aug-1000-hint --subsample-size 1000 --num-aug 1000 --hint
- Downloads last month
- 65