mattwesney/ToT_Reasoning_Problem_Solving_Dataset_V2

Train of Thought: Reasoning and Problem Solving V2

Synthetic English question–answer pairs with written reasoning for multidisciplinary problem solving. The dataset uses the creator’s “Train of Thought” framing, inspired by o1-style reasoning.

Contents

Rows: 72,117
Format: English text; JSON records with question, answer, id, and metadata (difficulty, topic, and generated reasoning).

Topics: Science, mathematics, engineering, computing, philosophy, logic, common sense, and practical problems.

Full curriculum.

Creation and limitations

The Genesis II pipeline generated the pairs using Qwen, Phi, and Llama models with structured prompting and format checks.

Generated responses can contain errors, bias, and uneven coverage. Written reasoning is not a guarantee that an answer is correct; evaluate samples for the intended use.

Use

from datasets import load_dataset

dataset = load_dataset("mattwesney/ToT_Reasoning_Problem_Solving_Dataset_V2")

Request access and authenticate if the dataset is gated.

License: MIT. Version: 2.0.0.

Citation

Matthew Wesney (2025), Train of Thought: Reasoning and Problem Solving V2. Hugging Face dataset.