The ML.ENERGY Leaderboard

Source code for The ML.ENERGY Leaderboard, which is a web leaderboard that displays results of The ML.ENERGY benchmark.

Running the Leaderboard Web App

Building the Data

After running the benchmark and collecting results, you can build the data for the leaderboard using the following command. This depends on The ML.ENERGY Data Toolkit, which can be installed with pip install mlenergy-data or used with uv run as shown below.

uv run --with mlenergy-data scripts/build_data.py \
  --output-dir public/data

If you are compiling the leaderboard data from a local directory that contains the ML.ENERGY Benchmark data, add --mlenergy-data-dir /path/to/compiled/data to the command above. In case you ran the benchmark on your own and have raw result data directories, modify the script to use LLMRuns.from_raw_results.

Web App Preview

npm install
npm run dev

Citation

@inproceedings{mlenergy-neuripsdb25,
    title={The {ML.ENERGY Benchmark}: Toward Automated Inference Energy Measurement and Optimization}, 
    author={Jae-Won Chung and Jeff J. Ma and Ruofan Wu and Jiachen Liu and Oh Jun Kweon and Yuxuan Xia and Zhiyu Wu and Mosharaf Chowdhury},
    year={2025},
    booktitle={NeurIPS Datasets and Benchmarks},
}