Math SFT Solutions No CoT V4: Large-Scale Mathematics Instruction Dataset

Version 4 of this dataset expands its scale and improves supervision quality through stronger cleaning and deduplication. It is a large-scale mathematics supervised fine-tuning dataset designed for instruction tuning and mathematical capability adaptation. The dataset was created by kaushik-harsh-99 and last updated on June 11, 2026.

Use Cases

• Instruction tuning for language models based on mathematical instruction-response pairs.
• Supervised fine-tuning for mathematical capability adaptation as described.
• Training models on direct instruction-to-response supervision without chain-of-thought reasoning.

Strengths

• Dataset scale was expanded in Version 4.
• Supervision quality was improved through stronger cleaning, deduplication, and formatting refinement.
• The dataset focuses on direct instruction-to-response supervision.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-06-11 10:35:16; freshness should be verified.

Provenance

Source

huggingface

Collection Method

Likely compiled from mathematical problem sources for supervised fine-tuning.

Freshness

Last updated 2026-06-11 10:35:16.