STEM Chain-of-Thought Supervised Fine-Tuning Dataset

Logics-STEM-SFT-Dataset-2.2M is a curated dataset for STEM reasoning, containing long Chain-of-Thought (CoT) examples for supervised fine-tuning. It is built from high-quality open-source data and enhanced through a curation and distillation process. The dataset is authored by Logics-MLLM and was last updated in January 2026.

Use Cases

• Supervised fine-tuning of language models for STEM reasoning using Chain-of-Thought examples.
• Training models on English language mathematics and science problems with step-by-step reasoning.
• Enhancing model performance on STEM tasks through curated long-form reasoning data.

Strengths

• Dataset is specifically curated for STEM reasoning tasks.
• Contains long Chain-of-Thought examples for supervised fine-tuning.
• Built from high-quality open-source data and enhanced through a rigorous process.
• Last updated in January 2026, indicating recent maintenance.

Limitations

• Specific row count, column structure, and file formats are unknown.
• Sample data is unavailable for inspection prior to download.
• License information is not provided, which may restrict usage.

Provenance

Source

Logics-MLLM

Collection Method

Curated from high-quality open-source data and enhanced through a rigorous curation and distillation process.

Freshness

Last updated on 2026-01-19.