📑 Paper    |    🌐 Project Page    |    💾 Released Resources    |    📦 Repo

This is the resource page of the our resources collection on Huggingface, we highlight your currect position with a blue block.

While having full executable code theoretically allows us to generate reliable execution trajectories as responses, two challenges arise: 1) Obtaining a deterministic reverse function for input prediction is impractical; 2) Automatically constructed trajectories are constrained by pre-designed templates and lack the expressiveness and generalizability of free-form natural language reasoning. Thus, we adopt a fully LLM-based approach for synthesizing all the desired responses using DeepSeek-V2.5, as it has top-tier performance but extremely low cost compared to other advanced LLMs.

*Due to our collaborators' compliance requirements, we only release the PythonEdu-Rs subset (this page) of full dataset.

The license for this dataset is ODC-BY.