Microsoft has recently introduced the BitNet b1.58 2B4T model, a groundbreaking development capable of running on various CPUs, including Apple's M2 chip. This model boasts an impressive 2 billion parameters and employs a low-precision architecture, achieving a compact memory footprint of merely 0.4GB. Despite its efficiency, BitNet b1.58 2B4T delivers performance on par with full-precision models of similar scale, while providing significant advantages in terms of energy consumption and reduced decoding latency.
