GSM8K Multiturn: Grade School Math Word Problems Reformatted as Multi-Turn Dialogues

A reformatted version of the GSM8K dataset of grade school math word problems. It presents the original 'socratic' version, where a model reflects by asking sub-questions, as a multi-turn conversation where the sub-questions are posed by a user. The dataset was created by author 'euclaise' and was last updated on Hugging Face in July 2024.

Use Cases

• Training conversational agents for multi-step reasoning based on the described dialogue structure.
• Evaluating the ability of language models to follow and answer a chain of sub-questions in a math context.
• Benchmarking model performance on mathematical reasoning presented as interactive dialogue.
• Studying the pedagogical effectiveness of the 'socratic' questioning method in AI-assisted learning.

Strengths

• Builds upon the established GSM8K benchmark for grade school math reasoning.
• Introduces a multi-turn conversational format for a known reasoning task.
• Dataset was last updated on 2024-07-06, suggesting recent maintenance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Reformatted from the 'socratic' version of the GSM8K dataset.

Freshness

Last updated 2024-07-06 18:52:23; freshness should be verified.