Competitive Programming with Large Reasoning Models

arxiv:2502.06807

Original source · Paper JSON

Reasoning and the "physics" of language models · 2025-02-03

Cites (0) · Cited by (1)

Complete cleaned Markdown · Source Markdown including bibliography

Read in parts

1. Part 1
2. Part 2
3. Part 3

Contents

• Competitive Programming with Large Reasoning Models
• Abstract
• 1 Introduction
• 2 OpenAI o1
• 2.1 CodeForces Benchmark
• 3 OpenAI o1-ioi
• 3.1 Coding RL Fine-tuning
• 3.2 o1-ioi Test-time Strategy
• Problem formulation
• Clustering
• Reranking
• Submission
• 3.3 CodeForces Benchmark
• 3.4 IOI 2024 Live Competition
• 4 OpenAI o3
• 4.1 CodeForces Benchmark
• 4.2 IOI 2024 Benchmark
• Sampling Approach.
• Results.
• 5 Software Engineering Evaluations
• 5.1 HackerRank Astra
• 5.2 SWE-Bench Verified
• 6 Conclusion
• Appendix A Authorship, credit attribution, and acknowledgments
• Data Preparation:
• IOI Model Training:
• Sampling Infrastructure:
• Test-time Strategy:
• Evaluation:
• Leadership:
• o3 Model Development
• Acknowledgments:
• Appendix B Additional CodeForces Details
• B.1 Data
• B.2 Grading
• B.3 Thinking Time
• B.4 Estimated Rating
• B.5 Percentile performance
• B.6 Per Problem Breakdown
• Appendix C IOI Submissions
• C.1 Nile
• C.2 Message
• C.3 Tree
• C.3.1 Submission 1
• C.3.2 Submission 2
• C.4 Hieroglyphs
• C.4.1 Submission 1
• C.4.2 Submission 2
• C.5 Mosaic
• C.5.1 Submission 1
• C.5.2 Submission 2
• C.6 Sphinx
• C.6.1 Submission 1
• C.6.2 Submission 2