BERT-Based Semantic Entropy under Landauer’s Principle — Dataset and Code

This dataset/code package reproduces all results for the paper “BERT-Based Semantic Entropy under Landauer’s Principle”. Contents include scripts to compute S_sem, plotting utilities, figures, and processed CSVs. How to use: (1) install requirements; (2) run semantic_entropy.py on your sentences; (3) generate E_norm with eta in [0.05, 0.10]; (4) use notebooks for plots. Please cite dataset DOI 10.6084/m9.figshare.30352402 and the preprint DOI 10.6084/m9.figshare.30352399. License: CC BY 4.0.