Status: Active (ongoing improvements and maintenance)
Kristian Virtanen 11th Jan 2025 [email protected]
This repository presents a reshaped version of the original Grade School Math (GSM8K) dataset, tailored to meet modern AI standards for reasoning and communication. While the questions remain unchanged from the original dataset, the answers have been rewritten to demonstrate logical reasoning and clarity, ensuring responses align with the expectations of today’s advanced language models.
The original GSM8K dataset, released by OpenAI, provided an excellent foundation for testing and training mathematical reasoning in large language models. However, the solutions lacked the step-by-step reasoning and modern conversational style that have become essential for effective AI responses. This updated dataset addresses these shortcomings, making it an invaluable resource for training models that excel in both problem-solving and communication.
Original Question:
Natalia sold clips to 48 of her friends in April, and then she sold half as many clips in May. How many clips did Natalia sell altogether in April and May?
Old Answer:
Natalia sold 48/2 = 24 clips in May.
Natalia sold 48+24 = 72 clips altogether in April and May.
Updated Answer:
### NEW ANSWER:
Let's break down Natalia's clip sales step by step to find out how many she sold in total over April and May.
1. **April Sales**: In April, Natalia sold clips to 48 friends. This means she sold **48 clips**.
2. **May Sales**: In May, she sold half as many clips as she did in April. To find out how many that is, we calculate:
\\( \\frac{48}{2} = 24 \\) clips sold in May.
3. **Total Sales**: Now, to find out how many clips Natalia sold altogether in both months, we simply add her April and May sales:
\\( 48 + 24 = 72 \\).
Therefore, Natalia sold a total of **72 clips** in April and May combined!
-
Enhancing Clarity and Reasoning:
Solutions now include a detailed, step-by-step breakdown of the reasoning process, ensuring that models trained on this dataset learn how to explain their answers in an accessible and transparent way. -
Modern AI Standards:
Answers have been rewritten to match the conversational and logical tone expected from modern language models like GPT-4. This ensures the dataset is more aligned with the capabilities and demands of today’s AI systems. -
Open Source Contribution:
This project breathes new life into the GSM8K dataset by making it more versatile for researchers, educators, and developers building cutting-edge models.
- Questions: The original questions remain untouched, preserving the high-quality, linguistically diverse grade school math problems created by human writers.
- Answers: All solutions have been rewritten to:
- Follow a clear, structured format.
- Provide step-by-step reasoning for solving each problem.
- Use conversational and approachable language without unnecessary embellishments.
- High-Quality Solutions: Each solution is crafted to provide detailed reasoning, fostering better understanding and stronger AI performance.
- Maintains Original Problem Integrity: Questions have been preserved to ensure compatibility with previous research benchmarks.
- Step-by-Step Explanations: Enables models to learn multi-step reasoning instead of just spitting out the final answer.
The dataset can be used for:
- Training AI models for multi-step mathematical reasoning.
- Evaluating a model’s ability to communicate solutions clearly and logically.
- Educational purposes, as a tool for teaching problem-solving skills.
This project builds upon the original GSM8K dataset by OpenAI. All credit for the question design and structure goes to the original authors. This updated version is our contribution to ensure the dataset stays relevant for modern AI research and applications.
For the original GSM8K dataset, visit the GSM8K GitHub repository.
This dataset is made available under the same licensing terms as the original GSM8K dataset. Please see the original license file for details.
Contributions are welcome! If you spot errors or have suggestions, feel free to submit a pull request or open an issue. Let’s keep improving together! 😊