Package Hallucination in LLM-Generated Python Code: Data and Code

Data and code for the study "Package Hallucination in LLM-Generated Python Code: Effects of Prompt Language and Region-Specific Tasks". The study asked three open-weight language models (gpt-oss-20b and gpt-oss-120b via the Groq API, and Gemma 4 26B A4B via the Gemini API) to write Python code for 243 tasks: 40 general, 40 specialized and 163 region-specific tasks covering ten countries and sixteen task types. Each task was given in English and in Turkish. Package names were extracted from the answers, checked against PyPI as of the date of each answer, and every name that did not exist was reviewed manually. The archive contains the task set, all model answers (collected 8 October to 3 November 2026), the extracted package mentions with their PyPI status, the manual review verdicts, task popularity counts from Stack Overflow, the analysis scripts and the data-collection notebooks. See README.md for details. Warning: some package names in these files were invented by the models and were not registered on PyPI when the data were collected. Do not install them.