QuantaSparkLabs/colab-strong-benchmark-reasoning

---
license: apache-2.0
task_categories:
- question-answering
- text-classification
language:
- en
size_categories:
- 100K<n<1M
---

# Colab Strong Benchmark: Reasoning & Trap (1.15GB Version)

A large-scale reasoning benchmark dataset with 2,875,000 carefully crafted questions containing logical traps and challenges.

## ☕ Dataset Statistics

### 📊 Dataset Visualizations

| Topic | Plot |
|---|---|
| **Category Distribution** | <img src="https://huggingface.co/datasets/QuantaSparkLabs/colab-strong-benchmark-reasoning/resolve/main/images/category_distribution.png" width="400"/> |
| **Difficulty Distribution** | <img src="https://huggingface.co/datasets/QuantaSparkLabs/colab-strong-benchmark-reasoning/resolve/main/images/difficulty_distribution.png" width="400"/> |
| **Question Length Distribution** | <img src="https://huggingface.co/datasets/QuantaSparkLabs/colab-strong-benchmark-reasoning/resolve/main/images/word_count_distribution.png" width="400"/> |
| **Trap Inclusion Distribution** | <img src="https://huggingface.co/datasets/QuantaSparkLabs/colab-strong-benchmark-reasoning/resolve/main/images/trap_inclusion_distribution.png" width="400"/> |

- **Total Samples**: 2,875,000
- **Total Size**: 1.164 GB
- **Number of Shards**: 58
- **Categories**: logical_reasoning, mathematical, code_debugging, adversarial_qa, contextual_reasoning, syllogisms, temporal_reasoning, counterfactual
- **Difficulty Levels**: easy, medium, hard, expert
- **Generated**: 2026-01-28

## ❖ Categories

- **logical_reasoning**: Logical challenges, puzzles, and reasoning problems
- **mathematical**: Logical challenges, puzzles, and reasoning problems
- **code_debugging**: Logical challenges, puzzles, and reasoning problems
- **adversarial_qa**: Logical challenges, puzzles, and reasoning problems
- **contextual_reasoning**: Logical challenges, puzzles, and reasoning problems
- **syllogisms**: Logical challenges, puzzles, and reasoning problems
- **temporal_reasoning**: Logical challenges, puzzles, and reasoning problems
- **counterfactual**: Logical challenges, puzzles, and reasoning problems

## ❦ Structure

Each sample contains:
- `id`: Unique identifier (q_00000000)
- `category`: Question category
- `difficulty`: easy/medium/hard/expert
- `question`: The full question text with context
- `options`: Multiple choice options (3-6 options)
- `correct_option`: Index of correct option (0-based)
- `explanation`: Step-by-step solution
- `tags`: Additional classification tags
- `metadata`: Complexity scores, trap indicators, timing estimates

## ➖ Usage

```python
from datasets import load_dataset

# Load in streaming mode (recommended)
dataset = load_dataset("QuantaSparkLabs/colab-strong-benchmark-reasoning", streaming=True)

# Or load a specific shard directly
import pandas as pd
df = pd.read_parquet("https://huggingface.co/datasets/QuantaSparkLabs/colab-strong-benchmark-reasoning/resolve/main/data/train-00000-of-00058.parquet")

# Sample usage for benchmarking
for example in dataset["train"].take(5):
print(f"Q: {example['question'][:100]}...")
print(f"A: Option {example['correct_option']}")
print()
```

## ➓ Generation

This dataset was programmatically generated with:
- **Rich text expansion** for natural language variety
- **Difficulty scaling** across multiple dimensions
- **Trap inclusion** (30% of questions contain intentional pitfalls)
- **Size optimization** for efficient loading and processing

## ➗ Performance Notes

- **Expected size**: ~1.15 GB
- **Actual size**: 1.164 GB
- **Compression**: Snappy compression applied to Parquet files
- **Loading**: Use streaming mode for memory-efficient processing

## ➘ Technical Details

- **Format**: Apache Parquet with Snappy compression
- **Sharding**: 58 shards of ~50,000 samples each
- **Columns**: 10 main columns with nested metadata
- **Language**: English
- **License**: Apache 2.0

---
*Built with ❤ by QuantaSparkLabs*
- **Overall Dataset Rating**: 9/10