microsoft/BitNet

★41,274C++🇺🇸 North AmericaMIT⑂ 3,844Issue 217Updated Jul 27, 2026

View on GitHub ↗

Official inference framework for 1-bit LLMs

CPU-first 1-bit LLM inference now extends to embeddings and ASR — Official framework for running 1-bit (BitNet) LLMs. Recent additions bring I2_S-quantized embedding models (0.6B/270M) with up to 2.28x CPU speedup over F16, plus a spin-off real-time multilingual ASR engine. Useful for developers wanting fast, low-bit inference on CPUs without GPUs.

Metrics

Retention can be computed 30 days after a repository first trends (it isn't shown until then). See About for how each metric is defined.

Star growth and Trending appearances (last 30 days)

Created Aug 5, 2024 · First on trending Oct 13, 2026

Other notable repositories in the same category (LLMs & ML)

• pst2154/nemotron-deciffusionStructured decisions from an 8B diffusion model in a single GPU pass
• todesking/slopsamplerCataloging the telltale phrases of AI-generated Japanese text
• datalab-to/indraA 1M-parameter OCR model that beats PaddleOCR and Tesseract on CPU
• fudan-generative-vision/RemorySoft memory that survives context compaction for LLM agents
• google-pytorch/torch_tpuA PyTorch Backend That Runs Models Directly on Google TPUs

See all LLMs & ML repositories →