April 16, 2025:
Microsofts Ultra-Efficient AI Model Runs on CPUs - Microsoft researchers unveiled BitNet b1.58 2B4T, a highly efficient 1-bit AI model designed to run on CPUs such as Apple's M2. BitNets achieve increased memory and computing efficiency by quantizing weights to -1, 0, and 1. With 2 billion parameters and training on 4 trillion tokens, the model outperforms traditional peers in benchmarks like GSM8K and PIQA, offering twice the speed with reduced memory usage.
However, the Microsoft bitnet.cpp framework, which is essential for this high performance, is limited to specific hardware, excluding GPUs. This presents compatibility challenges, hindering broader adoption of the technology across different systems.
