Sub-1-Bit LLM Compression via Latent Factorization

Researchers at Samsung Labs have developed a novel method for compressing large language models (LLMs) called Sub-1-Bit LLM Compression via Latent Factorization. The technique uses a latent factorization approach to reduce the size of LLMs while maintaining their performance. The researchers have open-sourced their implementation on GitHub, providing an example of how to use the method with the popular Hugging Face Transformers library.