High-Density Reasoning Samples in Pre-Training and Few-Shot GSM8K Accuracy

This report synthesises findings from 13 peer-reviewed papers addressing the following research question: How does varying the ratio of high-density reasoning samples in pre-training sequences affect few-shot accuracy on the GSM8K benchmark compared to uniform document mixing. 9 claims were extracted from source literature; 9 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 9.2/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research goal: How does varying the ratio of high-density reasoning samples in pre-training sequences affect few-shot accuracy on the GSM8K benchmark compared to uniform document mixing?

Autonomous literature synthesis. Automated review score: 9.2/10. Full text and citation available at Assignee Research.