The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits

Introduction of BitNet b1.58, a 1-bit Large Language Model variant, matches full-precision models in performance while being more cost-effective, paving the way for efficient LLM training and hardware design.