ELANA: A Simple Energy and Latency Analyzer for LLMs

Arxiv
Hung-Yueh Chiang, Bokun Wang, Diana Marculescu
The University of Texas at Austin
Report   Code  


🤗 Support models on Hugging Face, including Transformers, SSMs, and hybrid models
🔧 Support energy, latency, model size, KV cache size profiling
🔍 Support detailed kernel latency profiling


Compare to Zeus profiling framework


Hugging Face interface


Profile model size


Profiling results

Profile models on A6000


Profile models on Jetson series


Profiling kernels

We use torch profile to generate a json trace file and visualize it with Perfetto.


Citation

@article{chiang2025elana,
  title = {ELANA: A Simple Energy and Latency Analyzer for LLMs},
  author = {Chiang, Hung-Yueh and Wang, Bokun and Marculescu, Diana},
  journal = {arXiv preprint arXiv:2512.09946},
  year = {2025},
}


Acknowledgements

This work was supported in part by the ONR Minerva program, NSF CCF Grant No. 2107085, iMAGiNE - the Intelligent Machine Engineering Consortium at UT Austin, UT Cockrell School of Engineering Doctoral Fellowships.