Back to glossary

Term

llama.cpp

llama.cpp is an open-source C++ library for efficient LLM inference on CPU and GPU — the engine many local tools like Ollama and LM Studio are built on.

llama.cpp — explained in more detail

llama.cpp is a lean library written in modern C++ for local model inference. The project started in 2023 as a re-implementation of Meta’s LLaMA models without dependency on Python, PyTorch or CUDA — and quickly became the de-facto-standard engine for offline LLMs.

GGUF — the format that came with it

The move to GGUF introduced a dedicated model format that stores quantised weights efficiently. Quantisation — in short: reducing parameter precision from 16 or 32 bits to 4 or 8 bits — shrinks models to a fraction of their original size with surprisingly little quality loss.

Indirect relevance

llama.cpp is rarely used directly — it is the engine room. When someone starts a model with Ollama or LM Studio, llama.cpp is running underneath; when a tool offers a GGUF model, it comes from this library. That makes llama.cpp the implicit benchmark for the speed, hardware compatibility and model selection of the entire local LLM world.

Entdecke mehr

Themenuebersicht