<ul><li>Training strong coding models with RL has been hindered by the lack of high-quality datasets.</li><li>The DeepCoder team created a new training set of 24,000 validated problems by combining hand-curated problems.</li><li>Existing resources such as APPS, TACO, CodeContests, KodCode, and LeetCode are not suitable for RL benchmarks.</li><li>A custom filtering pipeline was used to ensure the quality of each problem in the dataset.</li></ul>

DeepCoder-14B-Preview: Open-Source Coding AI That Rivals Proprietary Models

Discover more