⬢github C++ · 40.9K ★ +157 since we first saw it · pushed 77 d ago · MIT
microsoft/BitNet
Official inference framework for 1-bit LLMs
bitnet.cpp is Microsoft's official C++ inference framework for 1-bit large language models (BitNet b1.58 style, ternary weights). It runs these ultra-low-bit models efficiently on CPUs (x86 AVX2, ARM NEON) and more recently GPUs, with optimized kernels and tools to convert models to compact I2_S quantization, enabling fast, lightweight local inference.
Why now: Recently Microsoft released the first 1-bit embedding models (BitNet-embedding-0.6B/270M) and a related real-time CPU ASR engine (VibeASR.cpp), keeping momentum behind efficient 1-bit inference.
Who it is for: Developers and researchers running or building efficient LLM/embedding inference locally on CPUs or edge devices.
Stars over our 43 snapshots: 40.7K to 40.9K, since 11 h ago.
Where people talked about it
- ⬢github microsoft/BitNet 8 min ago
API: https://socialmediatrends-api.osmike.com/v1/repos/microsoft/BitNet