You can compress LLMs for SE tasks to 1/49th their original size with minimal accuracy loss—making them practical to deploy while cutting environmental impact dramatically.
This paper presents Carbon-Taxed Transformers (CTT), a compression pipeline that makes large language models smaller, faster, and greener for software engineering tasks.