Our method of compressing large language models was adopted at NAACL, one of the largest AI conferences. The HIGGS method compresses LLM on phones and laptops while maintaining the quality of models. Experiments have shown that HIGGS compresses language models better than **** ogues.

The method was developed by scientists from the Yandex Research laboratory together with HSE, MIT, KAUST and ISTA. He's put in public on Hugging Face. : https://huggingface.co/doc...

You can read more about HIGGS in a scientific article on arXiv: https://arxiv.org/pdf/24