Microsoft has open-sourced bitnet.cpp, its official inference framework for 1.58-bit LLMs whose weights are restricted to three values (-1/0/+1), under the MIT license. Aimed at running large models on CPUs alone without a GPU, official benchmarks show up to 6.17x faster inference and up to 82.2% lower energy use.
Continue reading
The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in✓ Signed in — this article isn’t included in your current plan.