Neutral, reproducible data for the CPT→SFT case study in
https://github.com/anisiraj/llm_kickstart_repo (handbook.html → 🧪 Case Study).
corpus/*.txt — 41 deduplicated Wikipedia articles (comp/quantum chemistry), NFKC-normalized,
boilerplate stripped, equations preserved as inline $LaTeX$ (1,149 equations). ~112k tokens.
Source: English Wikipedia, CC BY-SA 4.0 (attribution required).
sft/seed_qa.jsonl — 32… See the full description on the dataset page:
https://huggingface.co/datasets/anisiraj/comp-chem-quantum-chem.
