Nemotron-Competitive-Programming-v1 is a large-scale synthetic programming and reasoning dataset released by NVIDIA in 2025, containing approximately 3.9 million samples with a total disk size of about 178 GB. It covers competitive programming problems in both Python and C++ as well as a cross-domain InfiniByte subset. The related research paper is "OpenCodeReasoning: Advancing Data Distillation for Competitive Coding", designed for supervised fine-tuning (SFT) tasks in code completion and code review, supporting the community's continued improvement of open-source models' capabilities in code writing and review.

Dataset Composition

• Competitive Coding (Python): Contains approximately 1.82 million samples, covering 34,799 unique competitive programming problems.
• Competitive Coding (C++): Contains approximately 930,000 samples, also based on the aforementioned competitive programming problems.
• InfiniByte: Contains approximately 1.17 million samples. This is a cross-domain subset that generates new programming problems by injecting concepts from scientific fields such as mathematics, physics, and chemistry into competitive programming problems.