This dataset accompanies the NeurIPS 2025 DL4C Workshop paper“Increasing LLM Coding Capabilities through Diverse Synthetic Coding Tasks”by Amal Abed, Ivan Lukic, Jörg K.H. Franke, and Frank Hutter.
It contains nearly 800k structured quadruplets of the form:
{ instruction, reasoning, solution_code, test_cases }
Each sample links a programming task with an explicit reasoning trace, a validated solution, and executable… See the full description on the dataset page:
https://huggingface.co/datasets/AutoML-org/SyntheticCode-800K.
