Multi-Domain Educational Synthetic Data For LLM Training

QVAC Genesis II expands the largest publicly available education-focused synthetic dataset for LLM training. It adds 10 new educational domains and introduces a novel Option-Level Reasoning Analysis data generation method, totaling 86 domains. The dataset was created by 'qvac' and updated in December 2025.

Use Cases

• Pre-training LLMs on multi-domain educational content using the 86 distinct subject areas.
• Conducting reasoning-centric post-training via the Option-Level Reasoning Analysis method.
• Benchmarking model performance across different educational domains present in the synthetic data.
• Studying the effect of synthetic educational data on model generalization and reasoning skills.

Strengths

• Expands upon Genesis I, described as the largest publicly available education-focused synthetic dataset.
• Adds 10 new educational domains, totaling 86 domains.

Limitations

• Specific row count, column details, and sample data are unavailable for assessment.
• The dataset is synthetic, which may not fully capture the complexity and nuance of real-world educational data.

Provenance

Source

huggingface

Collection Method

Synthetic data generation, including a novel Option-Level Reasoning Analysis method.

Freshness

Last updated in December 2025.