# Licenses and attribution

Independent teaching code in this directory is distributed under Apache-2.0 (LICENSE-APACHE-2.0.txt). prepare.py/scoring.py and protocol language reuse author code from article088, public commit4ed2968d7ca40b222d419372728a888529482e43; modules are included, no hidden dependency.

Transformers4.49.0 official commit a22a4378d97d06b7a1d9abad6e0086d30fdea199: Apache-2.0; original license retained verbatim. Executables use installed APIs, no vendored upstream Python. SmolLM2 model/tokenizer12fd25f77366fa6b3b4b768ec3050bf629380bac: model card Apache-2.0; downloaded, not redistributed. PyTorch2.6.0 installed official package: BSD-style upstream license; no copied source bundled. See pinned official LICENSE link in SOURCE_MAP.

Stanford SST-2 dataset8d51e7e4887a4caaa95b3fbebbf53c0490b58bbb card explicitly marks license unknown. Raw parquet, original review strings, prompts containing reviews and reversible input tokens are excluded. Public results contain indices, labels, lengths, masks, hashes and model-generated short sentiment responses. Download exact resources from publisher to reproduce; do not infer data rights from source-code license. Weights and generated fine-tuned state_dict remain local, excluded from public package.