The codebase for analysis and datasets used in position paper "Enough of Scaling LLMs! Lets focus on Downscaling" accepted at ICML 2025.
The codebase uses the scripts and datasets shared in https://github.com/SotaroKaneda/MLCarbon.
If you find our work useful in your projects/research, kindly cite our paper:
@inproceedings{
sengupta2025position,
title={Position: Enough of Scaling LLMs! Lets Focus on Downscaling},
author={Ayan Sengupta and Yash Goel and Tanmoy Chakraborty},
booktitle={Proceedings of the $\mathit{42}^{nd}$ International Conference on Machine Learning},
year={2025},
url={https://openreview.net/forum?id=CYJlJgEzZs}
}