21E12

We release the raw data for our processed PythonEdu-Rs dataset, adopted from the original dataset from HuggingFaceTB team. The data format for each line in the 0368500filteredv2ds25.sced.jsonl is as follows: Some of the ios are empty. The reason is that when executing the code, the input/output sizes are too large and exceed our required constraints. Thus, they are not stored or used later. Note: Due to imperfect LLM-based transformations, some problem descriptions do not contain enough information to describe the code. We leave this as future work to further enhance our data and update it to a better version.

This is the resource page of the our resources collection on Huggingface, we highlight your currect position with a blue block. Dataset Please also check the raw data after our processing if you are interested: DSACZ12DSA/Annoy-PyEdu-Rs-Raw. Models Introduction 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…