llama.cpp：让前沿 AI 完全本地运行的开源方案

2026-08-12 15:01· 29分钟前· kristianpaul

AI 导读

llama.cpp 让前沿 AI 完全运行在本地设备上，无需 API 密钥、无遥测、无限制，用户拥有模型与对话数据。它针对各类硬件优化，从笔记本到集群均可用同一二进制文件运行。现已支持 Qwen 3.6、Gemma 4、GPT-OSS 等模型，并可搭配本地编码智能体 Pi 使用。

原文 · 未翻译

AI that lives on your computer. Open-source, private & always local.

Run frontier AI entirely on your machine. No API keys, no telemetry, no limits. Own your models and conversation data.

curl -LsSf https://llama.app/install.sh | sh

Prefer Brew or Winget?

Package managers

Rather build from source?

Follow instructions

Pair it with a local coding agent.

Run llama serve, install the pi-llama plugin and launch Pi. It will automatically discover your local model. No config, no API keys. Files stay on your machine, requests never leave it.


llama serve


pi install git:github.com/huggingface/pi-llama


pi

Optimized for any hardware.

From your laptop to a cluster, llama.cpp runs on whatever you have. Same binary, same models, same hand-tuned kernels for every GPU and CPU.

Apple Silicon

M Ultra

RTX 5090

CPU

Jetson

H100

MI300

RTX 4090

A100

M Pro

M Max

DGX Spark

Radeon RX

B200

Intel Arc

RTX 3090

Run your first model

Qwen 3.6

Alibaba's next-gen natively multimodal reasoning models. Dense and MoE variants that rival models many times their size on coding and vision tasks.

Gemma 4

Google's most capable open models, built from Gemini 3 technology. Supports multimodal reasoning, agentic workflows, and 140+ languages.

GPT-OSS

OpenAI's first open-weight models since GPT-2. Built for reasoning, agentic tasks, and developer use with function calling and tool use capabilities.

Gemma 3

Google's multimodal models built from Gemini technology. Supports 140+ languages, vision, and text tasks with up to 128K context for edge to cloud deployment.

Browse all models

来源：Hacker News 热门（buzzing.cc 中文翻译） · llama.app