F26 Data Privacy: Lab 1
Student materials for Tianhao Wang's Fall 2026 Data Privacy course at UVA.
Download the notebook, or use the copy provided through Canvas. Upload it to Google Colab and run Setup. Models and supporting files download automatically; no Drive mount or Hugging Face login is needed. CPU execution is sufficient. All tasks, grading criteria, and submission instructions are included in the notebook.
These small course GPTs support controlled classroom experiments with fictional
records and planted canaries. The practice experiment includes a fine-tuned
target and labeled examples.
They are not general-purpose language models or production privacy benchmarks.
Use the notebook's loader for these .pt weights, not Transformers AutoModel.
Only student materials are hosted here. Solutions, hidden test cases, and instructor training manifests are not included. The notebook pins a fixed revision so students use the same course files.
The GPT implementation is adapted from nanoGPT/minGPT and uses GPT-2 tokenizer files. Upstream notices are in THIRD_PARTY_NOTICES.md.