Installer llama.cpp avec Homebrew

Liens du projet

• Page d'accueil
• Dépôt

Commandes et alias

• llama
• llama-batched
• llama-batched-bench
• llama-bench
• llama-cli
• llama-completion
• llama-debug
• llama-debug-template-parser
• llama-diffusion-cli
• llama-embedding
• llama-eval-callback
• llama-finetune
• llama-fit-params
• llama-gen-docs
• llama-gguf
• llama-gguf-hash
• llama-gguf-split
• llama-idle
• llama-imatrix
• llama-lookahead
• llama-lookup
• llama-lookup-create
• llama-lookup-merge
• llama-lookup-stats
• llama-mtmd-cli
• llama-parallel
• llama-passkey
• llama-perplexity
• llama-quantize
• llama-results
• llama-retrieval
• llama-server

Historique du projet

The repository was created on GitHub on March 10, 2023, shortly after Meta's LLaMA model release changed the center of gravity for local language-model experimentation. The README states the project goal as LLM inference with minimal setup and strong performance across local and cloud hardware.

The project is closely tied to ggml. Its README describes llama.cpp as the main playground for developing new ggml features, and the implementation grew around plain C/C++, integer quantization, CPU backends, and hardware accelerators such as Metal, CUDA, Vulkan, SYCL, HIP, and related GPU paths.

As model support broadened beyond the original LLaMA family, llama.cpp became a runtime and tooling umbrella: converters, quantizers, benchmarking tools, embedding tools, an OpenAI-compatible server, multimodal support, and many model-family loaders are represented in the command set and documentation.

Historique d'adoption

Package adoption spread because llama.cpp lowered the cost of trying local inference: build from source, install from Homebrew, Nix, winget, conda-forge, Docker, or download release binaries, then run a model file or fetch one from Hugging Face-oriented workflows.

The README's bindings list shows the surrounding ecosystem that formed around the C/C++ core, including Python, Go, Node.js, Ruby, browser/Wasm, editor-completion plugins, and server clients. That ecosystem made llama.cpp both an end-user CLI and a library/runtime target for other packages.

Its high-frequency build-tag release pattern reflects active downstream pressure: package managers, bindings, model hubs, and local-AI applications all depend on fast propagation of backend, quantization, and model-format changes.

Modes d'utilisation

Users run llama-cli for local prompts, llama-server for an OpenAI-compatible HTTP API, llama-bench for performance testing, llama-quantize for smaller model files, and many auxiliary tools for embeddings, perplexity, retrieval, tokenization, and model-file manipulation.

The package is especially useful when a developer wants a self-contained inference engine: compile once, point it at a model, and choose a CPU/GPU backend without adopting a full ML framework stack.

Pourquoi les passionnés de paquets s'y intéressent

For package maintainers, llama.cpp is unusually dynamic: hardware backend flags, model-format transitions, CLI renames, bundled tools, and release cadence all matter. It turned local AI into something package managers had to treat like a fast-moving systems tool rather than a single Python application.

It is also a packaging bridge between model hubs and Unix tooling. The same project can be installed as a formula, used as a server daemon, linked by language bindings, wrapped by desktop apps, or embedded in another inference product.

Chronologie

• 2023: ggml-org/llama.cpp repository created on March 10.
• 2023: The project manifesto discussion documented goals and direction for the early community.
• 2024: llama.cpp issue trackers began maintaining separate API changelogs for libllama and llama-server.
• 2025: Multimodal support reached llama-server through the upstream pull request and documentation referenced from the README.
• 2026: Homebrew, Nix, winget, conda-forge, Docker, and release binaries were documented installation paths in the upstream README.

Related projects

• ggml is the closest related project, because llama.cpp is the main development playground for that tensor library.
• Important downstream and adjacent projects include llama-cpp-python, node-llama-cpp, go-llama.cpp, wllama, llama.vscode, llama.vim, and applications that wrap llama-server's OpenAI-compatible API.

Sources

• https://api.github.com/repos/ggml-org/llama.cpp
• https://formulae.brew.sh/formula/llama.cpp
• https://github.com/ggml-org/ggml
• https://github.com/ggml-org/llama.cpp
• https://github.com/ggml-org/llama.cpp/discussions/205
• https://github.com/ggml-org/llama.cpp/issues/9289
• https://github.com/ggml-org/llama.cpp/issues/9291
• https://github.com/ggml-org/llama.cpp/pull/12898
• https://github.com/ggml-org/llama.cpp/releases
• https://raw.githubusercontent.com/ggml-org/llama.cpp/master/README.md