A cleaned mathematical supervised fine-tuning dataset designed for instruction tuning and mathematical capability adaptation. The dataset introduces a simplified instruction–response format and removes intermediate reasoning contamination. It was created by author kaushik-harsh-99 and was last updated on 2026-06-07.
Use Cases
- Instruction tuning of language models based on the simplified instruction–response format.
- Adapting models for mathematical capability based on the dataset's mathematical content.
- Training models on GSM8K-style problems using the augmented mathematical responses mentioned.
- Evaluating model performance on tasks without chain-of-thought reasoning contamination.
Strengths
- Dataset is specifically cleaned for supervised fine-tuning, removing intermediate reasoning contamination.
- Version 2 introduces a simplified instruction–response format.
- Includes augmented mathematical responses generated over GSM8K-style problems.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- huggingface
- Freshness
- Last updated 2026-06-07 04:28:17; freshness should be verified.