AI Energy Score Leaderboard by Sasha Luccioni, Boris Gamazaychikov, Emma Strubell, Sara Hooker, Yacine Jernite, Margaret Mitchell and Scott Chamberlin.
Source repository license: Apache-2.0 (Hugging Face repository metadata).
Pinned revision: c37e01add3d7f295f4bd2a8af0293edc6bf25fb2
Original energy: https://huggingface.co/spaces/AIEnergyScore/Leaderboard/raw/c37e01add3d7f295f4bd2a8af0293edc6bf25fb2/data/energy/text_generation.csv
Original parameters: https://huggingface.co/spaces/AIEnergyScore/Leaderboard/raw/c37e01add3d7f295f4bd2a8af0293edc6bf25fb2/data/params/text_generation.csv
AI Footprint Lab includes every February 2025 text-generation Class A row whose parameter table explicitly lists one GPU. No energy-based selection. Sorted alphabetically.
Energy CSV total_gpu_energy is kWh per 1,000 queries: the original app multiplies by 1,000 to display Wh per 1,000 queries. AI Footprint Lab preserves both units.
Methodology: https://github.com/huggingface/AIEnergyScore/blob/a5b9a048ecd44aa087cdd5b575a17d705128c424/index.md
Method documentation is a separately pinned snapshot, not a verified record of each historical run configuration.
Retrieved 2026-09-05. AI Footprint Lab extraction has not received independent scientific review.

License copy: LICENSE.txt in this directory. Original CSV files are unmodified; AI Footprint Lab’s filtered and converted records are in src/data/reference/benchmarks.json.