Microsoft Research has unveiled '1-bit,' a groundbreaking two-billion parameter language model that operates efficiently on standard CPUs. Trained on 4 trillion tokens, this open-source model matches the performance of full-precision models while using 82.2% less energy and delivering 6.17x faster inference. The model requires only 0.4GB of memory, significantly less than comparable models needing 2-5GB, and functions effectively on both ARM and x86 processors, making advanced AI more accessible without specialized hardware.
Microsoft Research has unveiled '1-bit,' a groundbreaking two-billion parameter language model that operates efficiently on standard CPUs. Trained on 4 trillion tokens, this open-source model matches the performance of full-precision models while using 82.2% less energy and delivering 6.17x faster inference. The model requires only 0.4GB of memory, significantly less than comparable models needing 2-5GB, and functions effectively on both ARM and x86 processors, making advanced AI more accessible without specialized hardware.