Microsoft researchers have developed a hyper-efficient AI model called BitNet b1.58 2B4T that can run on CPUs, including Apple’s M2. This 1-bit model, also known as a “bitnet,” is the largest-scale bitnet to date with 2 billion parameters. Bitnets are compressed models designed to run on lightweight hardware.
They quantize weights into just three values: -1, 0, and 1, making them more memory and computing-efficient than most models today. BitNet b1.58 2B4T was trained on a dataset of 4 trillion tokens, equivalent to about 33 million books.
BitNet redefines AI efficiency
It outperforms traditional models of similar sizes, according to the researchers. The model surpasses Meta’s Llama 3.2 1B, Google’s Gemma 3 1B, and Alibaba’s Qwen 2.5 1.5B on benchmarks including GSM8K, a collection of grade-school-level math problems, and PIQA, which tests physical commonsense reasoning skills. BitNet b1.58 2B4T is also speedier than other models of its size, in some cases twice as fast, while using a fraction of the memory.
However, achieving that performance requires using Microsoft’s custom framework, bitnet.cpp, which only works with certain hardware at the moment. GPUs, which dominate the AI infrastructure landscape, are notably absent from the list of supported chips. The researchers state that bitnets may hold promise, particularly for resource-constrained devices, but compatibility remains a significant sticking point.
Image Credits: Photo by Ashkan Forouzani on Unsplash
Johannah Lopez is a versatile professional who seamlessly navigates two worlds. By day, she excels as a SaaS freelance writer, crafting informative and persuasive content for tech companies. By night, she showcases her vibrant personality and customer service skills as a part-time bartender. Johannah's ability to blend her writing expertise with her social finesse makes her a well-rounded and engaging storyteller in any setting.
