菜单
BitNet.cpp — official inference framework cho 1-bit LLMs: 1.37x-6.17x speedup, -55-82% energy. Chạy 100B model trên single CPU 5-7 tok/s. ARM + x86.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/yanacuti1121/Yana-AI --skill microsoft-bitnet命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | microsoft--bitnet |
| description | BitNet.cpp — official inference framework cho 1-bit LLMs: 1.37x-6.17x speedup, -55-82% energy. Chạy 100B model trên single CPU 5-7 tok/s. ARM + x86. |
| allowed-tools | Bash, Read, Write |
| user-invocable | true |
BitNet.cpp: inference framework cho 1.58-bit quantized LLMs — lossless, fast, energy-efficient. Chạy model 100B trên CPU thông thường.
git clone https://github.com/microsoft/BitNet
cd BitNet
pip install -r requirements.txt
# Build: cmake + make
# Inference
python run_inference.py \
-m models/BitNet-b1.58-2B-4T \
-p "What is AI?" \
-cnv
# Benchmark
python benchmark.py -m models/BitNet-b1.58-2B-4T --threads 4
# Convert SafeTensors → BitNet format
python convert_safetensors.py --input model.safetensors
ARM CPUs: 1.37x – 5.07x speedup | -55% – 70% energy
x86 CPUs: 2.37x – 6.17x speedup | -72% – 82% energy
100B model trên single CPU: 5-7 tokens/second
Microsoft:
BitNet-b1.58-2B-4T (2.4B params) — official flagship
Community:
BitNet 0.7B – 3.3B
Llama3-8B-1.58
Falcon3 1B – 10B
Falcon-E 1B – 3B
Kernels: I2_S, TL1, TL2 (platform-specific)
https://github.com/microsoft/BitNet · MIT · official inference framework