hendrycks/apps

Provides a large-scale benchmark dataset of 10,000+ competitive programming problems with input-output test cases for evaluating code generation models. Includes training and evaluation pipelines designed for fine-tuning transformer-based language models (GPT-2, GPT-Neo) on Python code synthesis tasks. Available via Hugging Face Datasets integration with pre-trained model weights and standardized metrics for measuring programming competence across difficulty levels.

520 stars. No commits in the last 6 months.