# System map — ggml-org/llama.cpp

> Canonical page: https://codevetter.com/unpack/llama-cpp/system-map

# System map — ggml-org/llama.cpp

Scanned commit 7ab4ee7baad2d920464cbacfad4f4b07cf111fd2 (2026-09-22). Generated by codevetter codevetter 1.12.0 on 2026-09-22.

The repo is a CMake C/C++ monorepo layered as: ggml/ (tensor ops + hardware backend plugins, statically registered or dlopen'd), src/ (libllama: GGUF model loading, vocab/tokenizers, KV-cache memory, compute graph builders per architecture, samplers, public C API in include/llama.h), common/ (llama-common utility lib: arg parsing, chat templates via a custom Jinja engine, sampling orchestration, HF downloading, subprocess helpers), tools/ (user-facing binaries incl. server, cli, mtmd multimodal, quantize, perplexity, bench), app/ (unified `llama` binary), and Python conversion tooling (conversion/, gguf-py/, convert_hf_to_gguf.py). Tests live in tests/ (CTest C++ binaries) and tools/server/tests (pytest). The web UI is a SvelteKit app in tools/ui embedded into the server binary at build time.
- The unified `llama` binary is a thin dispatcher: app/llama.cpp defines a static command table `cmds[]` mapping subcommands (serve, cli, update, download, completion, bench, batched-bench, fit-params, quantize, perplexity, version, licenses, help) to entry functions like llama_server and llama_cli, and sets the LLAMA_APP_CMD env var before invoking them so router child processes re-invoke correctly. — [app/llama.cpp#L64-78](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/app/llama.cpp#L64-L78) [app/llama.cpp#L129-143](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/app/llama.cpp#L129-L143)
- The public API is a C ABI in include/llama.h (~1646 lines, ~246 LLAMA_API declarations) implemented primarily by src/llama.cpp, which forwards into internal classes llama_model, llama_context, llama_vocab, and llama_memory. — [include/llama.h](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/include/llama.h) [src/llama.cpp#L1-30](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/src/llama.cpp#L1-L30)
- Model instantiation is a giant architecture switch: llama_model_mapping() in src/llama-model.cpp maps each LLM_ARCH_* enum to a `new llama_model_<arch>(params)` subclass (e.g. llama_model_llama, llama_model_deepseek2); unknown archs throw 'unsupported model architecture'. — [src/llama-model.cpp#L43-373](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/src/llama-model.cpp#L43-L373)
- Per-architecture graph builders live in src/models/ (156 files, e.g. llama.cpp, deepseek2.cpp, qwen3.cpp) plugged into llama-graph.cpp which assembles ggml compute graphs executed via llama_context::decode(). — [src/llama-graph.cpp](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/src/llama-graph.cpp) [src/llama-context.cpp#L4326-4335](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/src/llama-context.cpp#L4326-L4335) [src/llama-graph.h#L101-363](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/src/llama-graph.h#L101-L363)
- ggml is vendored in ggml/ with a backend registry: ggml_backend_registry's constructor statically registers compile-time backends (CUDA, Metal, SYCL, Vulkan, WebGPU, zDNN, VirtGPU, OpenCL, ZenDNN, Hexagon, CANN, BLAS, RPC, OpenVINO, ET, CPU) and ggml_backend_load()/ggml_backend_load_all() can dlopen backend shared libraries at runtime (GGML_BACKEND_DL). — [ggml/src/ggml-backend-reg.cpp#L115-174](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/ggml/src/ggml-backend-reg.cpp#L115-L174) [ggml/src/ggml-backend-reg.cpp#L393](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/ggml/src/ggml-backend-reg.cpp#L393) [ggml/src/ggml-backend-reg.cpp#L574-603](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/ggml/src/ggml-backend-reg.cpp#L574-L603)
- common/ builds the llama-common static library holding shared tool infrastructure: common_params + arg parsing (arg.cpp, ~4.7k lines), chat template/tool-call handling (chat.cpp + jinja/ + parsers/), sampler orchestration (sampling.cpp), speculative decoding (speculative.cpp), HF/URL downloading (download.cpp, hf-cache.cpp), subprocess wrapper (subproc.cpp) and a cpp-httplib HTTP client helper (http.h). — [common/arg.cpp#L77-84](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/common/arg.cpp#L77-L84) [common/chat.h#L316-340](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/common/chat.h#L316-L340) [common/sampling.cpp#L187-210](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/common/sampling.cpp#L187-L210) [common/download.cpp#L45-90](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/common/download.cpp#L45-L90) [common/http.h#L100-118](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/common/http.h#L100-L118)
- tools/server implements the OpenAI/Anthropic-compatible HTTP server on cpp-httplib (vendor/cpp-httplib): server-http.cpp wraps httplib::Server/SSLServer, server.cpp registers all routes, server-context.cpp owns the inference slots/queue, and server-models.cpp implements router mode by spawning child llama-server subprocesses. — [tools/server/server-http.cpp#L7-22](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/server/server-http.cpp#L7-L22) [tools/server/server.cpp#L241-306](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/server/server.cpp#L241-L306) [tools/server/server-models.cpp#L1163-1189](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/server/server-models.cpp#L1163-L1189) [vendor/cpp-httplib](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/vendor/cpp-httplib)
- tools/cli is an interactive chat client that embeds the server: cli_context::init() spawns llama_server() on a std::thread bound to a random free port (cli_server::start) and all generation goes through HTTP SSE to /v1/chat/completions rather than calling libllama directly. — [tools/cli/cli-server.h#L33-56](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/cli/cli-server.h#L33-L56) [tools/cli/cli-context.cpp#L94-151](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/cli/cli-context.cpp#L94-L151) [tools/cli/cli-context.cpp#L350-411](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/cli/cli-context.cpp#L350-L411)
- Multimodal support is a separate mtmd library in tools/mtmd: clip.cpp (~6k lines) implements the vision encoder stack, mtmd-audio.cpp handles audio input, and mtmd_init_from_file() loads a separate --mmproj GGUF projector model alongside the main model. — [tools/mtmd/mtmd.cpp#L582](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/mtmd/mtmd.cpp#L582) [tools/mtmd/mtmd.cpp#L1085](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/mtmd/mtmd.cpp#L1085) [tools/mtmd/clip.cpp](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/mtmd/clip.cpp)
- Model conversion is Python: convert_hf_to_gguf.py drives per-architecture converter classes in conversion/ registered via @ModelBase.register, writing GGUF through the gguf-py package (gguf-py/pyproject.toml, name=gguf v0.19.0); tensor name/layout tables live in gguf-py/gguf/constants.py. — [convert_hf_to_gguf.py#L15-25](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/convert_hf_to_gguf.py#L15-L25) [conversion/base.py](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/conversion/base.py) [gguf-py/pyproject.toml](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/gguf-py/pyproject.toml) [docs/development/HOWTO-add-model.md](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/docs/development/HOWTO-add-model.md)
- The web UI is a SvelteKit + Vite + Tailwind app in tools/ui (package.json name=llama-ui) that is compiled to static assets and embedded into the server binary via a CMake-generated ui.cpp; LLAMA_USE_PREBUILT_UI downloads prebuilt assets from the HF bucket 'ggml-org/llama-ui' by default. — [tools/ui/package.json](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/ui/package.json) [tools/ui/CMakeLists.txt#L1-37](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/ui/CMakeLists.txt#L1-L37) [tools/server/server-http.cpp#L5](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/tools/server/server-http.cpp#L5)
- Top-level CMakeLists.txt is the single build system (cmake 3.14...3.28): version is LLAMA_VERSION 0.4.1-dev, feature options include LLAMA_BUILD_COMMON/TESTS/TOOLS/EXAMPLES/SERVER/APP/UI, LLAMA_OPENSSL, LLAMA_SUBPROCESS, LLAMA_LLGUIDANCE, plus deprecated-option shims mapping LLAMA_CUDA etc. to GGML_* options. — [CMakeLists.txt#L1-23](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/CMakeLists.txt#L1-L23) [CMakeLists.txt#L129-146](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/CMakeLists.txt#L129-L146) [CMakeLists.txt#L174-195](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/CMakeLists.txt#L174-L195)
- Third-party code is vendored under vendor/ (cpp-httplib, nlohmann json, miniaudio, stb, sheredom subprocess.h, hash) with a dedicated CI workflow .github/workflows/check-vendor.yml verifying vendor integrity. — [vendor/CMakeLists.txt](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/vendor/CMakeLists.txt) [.github/workflows/check-vendor.yml](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/.github/workflows/check-vendor.yml)
- Deployment shapes: static or shared libllama (BUILD_SHARED_LIBS default ON except MinGW/Emscripten), the `llama` unified binary, individual llama-* tool binaries, a WASM/Emscripten build (MEMORY64 enabled), Docker images via docs/docker.md, and an install-script self-update path gated on LLAMA_INSTALL_BUILD. — [CMakeLists.txt#L55-79](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/CMakeLists.txt#L55-L79) [app/llama.cpp#L27-41](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/app/llama.cpp#L27-L41) [docs/build.md](https://github.com/ggml-org/llama.cpp/blob/7ab4ee7baad2d920464cbacfad4f4b07cf111fd2/docs/build.md)

Every claim links to its cited files at the pinned commit. Citations identify source evidence; they do not establish runtime correctness. Independent CodeVetter scan — not reviewed by upstream maintainers.

## Public product links

- [CodeVetter](https://codevetter.com/)
- [Download](https://codevetter.com/download)
- [Documentation](https://codevetter.com/docs/)
- [Source](https://github.com/Codevetter/codevetter)