| name | llamacpp-docs |
| description | Use when working with llama.cpp: building from source, installation, GGUF model loading, quantization, speculative decoding, function calling, multimodal models (LLaVA, Gemma3, MiniCPM), GPU backends (CUDA, Metal, Vulkan, SYCL, OpenCL, CANN), Android deployment, Docker setup, performance tuning, or adding new model architectures to llama.cpp. |
llama.cpp Documentation
Official llama.cpp docs (sourced from github.com/ggml-org/llama.cpp/docs).
CRITICAL: grep references/ for keywords before answering.
Topic Index
Build & Install
build.md โ Building from source (CMake)
install.md โ Installation guide
docker.md โ Docker setup
android.md โ Android deployment
build-riscv64-spacemit.md โ RISC-V 64-bit build
build-s390x.md โ s390x architecture build
Features
function-calling.md โ Function calling / tool use
speculative.md โ Speculative decoding
multimodal.md โ Multimodal model support (overview)
preset.md โ Preset configurations
llguidance.md โ LLGuidance grammar integration
autoparser.md โ Auto-parser for structured output
ops.md โ Supported operations reference
GPU Backends
backend/CUDA-FEDORA.md โ CUDA on Fedora
backend/SYCL.md โ Intel SYCL backend
backend/OPENCL.md โ OpenCL backend
backend/BLIS.md โ BLIS backend
backend/CANN.md โ Huawei Ascend CANN backend
backend/VirtGPU.md โ Virtual GPU backend
backend/VirtGPU/ โ VirtGPU configuration & development
backend/ZenDNN.md โ AMD ZenDNN backend
backend/zDNN.md โ IBM zDNN backend
backend/snapdragon/ โ Qualcomm Snapdragon NPU
Multimodal Models
multimodal/llava.md โ LLaVA vision-language model
multimodal/gemma3.md โ Gemma 3 multimodal
multimodal/glmedge.md โ GLM Edge model
multimodal/granitevision.md โ Granite Vision
multimodal/MobileVLM.md โ Mobile VLM
multimodal/minicpm*.md โ MiniCPM series (v2.5, v2.6, v4.0, v4.5, o2.6, o4.0)
Development
development/HOWTO-add-model.md โ Adding new model architectures
development/token_generation_performance_tips.md โ Performance optimization
development/debugging-tests.md โ Debugging tests
development/parsing.md โ Parsing internals