{"schemaVersion":1,"id":"llama-cpp","revision":"2026-09-15.1","name":"llama.cpp","monogram":"L","category":"Inference library","openness":"Open source","license":"MIT runtime source; model weights have separate licenses","color":"#b4a0ee","summary":"C/C++ inference library and server with GGUF models and selectable hardware backends.","website":"https://github.com/ggml-org/llama.cpp","repository":"https://github.com/ggml-org/llama.cpp","environment":"Self-managed local build and server; pin compiled backends and model files.","contributor":"Testedknowhow model runtime research","affiliation":"Commissioned by a Nimbalyst maintainer. Documentation synthesis; no independent review or product execution.","contentLicense":"Unlicensed prototype; redistribution decision pending","reviewStatus":"unreviewed","lastChecked":"2026-09-15","expiresAt":"2026-10-15","applicability":{"versions":["Repository revision 38a5b42d9a3e82e0a586bcd1caed121f36c87a73; living documentation separately dated."],"environment":"Self-managed local build and server; pin compiled backends and model files.","excludes":["Runtime verification and performance rankings","Model weights license or quality evaluation","Cloud inference, remote hosts and external tools","Unlisted model architectures, builds and hardware"]},"claims":[{"id":"llama-cpp-rt-local","capability":"rt-local","statement":"Supports inference on a local machine.","method":"documented","outcome":"claimed","scope":"Supported model and configured local backend; documentation only, no execution recorded.","evidenceIds":["llama-cpp-readme"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-apple-silicon","capability":"rt-apple-silicon","statement":"Documents Metal acceleration on Apple Silicon.","method":"documented","outcome":"claimed","scope":"Supported model and configured local backend; documentation only, no execution recorded.","evidenceIds":["llama-cpp-readme"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-cpu","capability":"rt-cpu","statement":"Documents CPU inference with architecture-specific optimizations.","method":"documented","outcome":"claimed","scope":"Supported model and configured local backend; documentation only, no execution recorded.","evidenceIds":["llama-cpp-readme"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-nvidia","capability":"rt-nvidia","statement":"Provides a CUDA backend for supported NVIDIA GPUs.","method":"documented","outcome":"claimed","scope":"Supported model and configured local backend; documentation only, no execution recorded.","evidenceIds":["llama-cpp-readme"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-gguf","capability":"rt-gguf","statement":"Loads supported models from GGUF files.","method":"documented","outcome":"claimed","scope":"Supported model and configured local backend; documentation only, no execution recorded.","evidenceIds":["llama-cpp-readme"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-quantized","capability":"rt-quantized","statement":"Supports multiple integer weight quantization formats.","method":"documented","outcome":"claimed","scope":"Supported model and configured local backend; documentation only, no execution recorded.","evidenceIds":["llama-cpp-readme"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-http","capability":"rt-http","statement":"Provides a local HTTP inference server.","method":"documented","outcome":"claimed","scope":"Self-managed server; configure bind address and access controls.","evidenceIds":["llama-cpp-server"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-openai","capability":"rt-openai","statement":"Exposes an OpenAI-style chat-completions endpoint.","method":"documented","outcome":"claimed","scope":"The server explicitly avoids promising full OpenAI API compatibility.","evidenceIds":["llama-cpp-server"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-streaming","capability":"rt-streaming","statement":"Streams completion output using server-sent events.","method":"documented","outcome":"claimed","scope":"The documented /completion stream option.","evidenceIds":["llama-cpp-server"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-structured","capability":"rt-structured","statement":"Accepts JSON schema constraints for generated responses.","method":"documented","outcome":"claimed","scope":"Supported schemas and model configuration; semantic correctness untested.","evidenceIds":["llama-cpp-server"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-concurrent","capability":"rt-concurrent","statement":"Supports parallel decoding and continuous batching.","method":"documented","outcome":"claimed","scope":"Configured server slots and available resources; no throughput measured.","evidenceIds":["llama-cpp-server"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-prompt-cache","capability":"rt-prompt-cache","statement":"Can reuse a matching prompt prefix through the KV cache.","method":"documented","outcome":"claimed","scope":"cache_prompt; reuse can affect determinism across batch sizes.","evidenceIds":["llama-cpp-server"],"contradictions":[],"support":"yes"},{"id":"llama-cpp-rt-speculative","capability":"rt-speculative","statement":"Exposes draft-model speculative decoding controls.","method":"documented","outcome":"claimed","scope":"Compatible draft setup required; acceleration is not guaranteed.","evidenceIds":["llama-cpp-server"],"contradictions":[],"support":"yes"}],"evidence":[{"id":"llama-cpp-readme","url":"https://github.com/ggml-org/llama.cpp/blob/38a5b42d9a3e82e0a586bcd1caed121f36c87a73/README.md","title":"llama.cpp overview","locator":"Description; supported backends; quick start","capturedAt":"2026-09-15","kind":"repository","revision":"38a5b42d9a3e82e0a586bcd1caed121f36c87a73","redistribution":"Link and original paraphrase only; upstream rights retained."},{"id":"llama-cpp-server","url":"https://github.com/ggml-org/llama.cpp/blob/38a5b42d9a3e82e0a586bcd1caed121f36c87a73/tools/server/README.md","title":"llama.cpp server documentation","locator":"Features; --model; /completion; /v1/chat/completions; cache_prompt; speculative decoding","capturedAt":"2026-09-15","kind":"repository","revision":"38a5b42d9a3e82e0a586bcd1caed121f36c87a73","redistribution":"Link and original paraphrase only; upstream rights retained."},{"id":"llama-cpp-license","url":"https://github.com/ggml-org/llama.cpp/blob/38a5b42d9a3e82e0a586bcd1caed121f36c87a73/LICENSE","title":"llama.cpp license","locator":"MIT License","capturedAt":"2026-09-15","kind":"repository","revision":"38a5b42d9a3e82e0a586bcd1caed121f36c87a73","redistribution":"Link and original paraphrase only; upstream rights retained."}],"gaps":["Which pinned model, quantization, context length and runtime versions fit the selected Mac?","What are measured latency, throughput, peak memory and swap under one and four concurrent requests?"],"invalidationTriggers":["Runtime, bundled engine, backend, model format or license changes.","Documentation changes affecting the cited configuration."],"verificationPlan":["Choose hardware and obtain explicit approval for installations, downloads and execution resources.","Pin runtime and engine versions, model artifact hashes, tokenizer, template, quantization and inference settings.","Run the local-model-mac investigation with synthetic prompts and capture raw measurements; do not record documentation as execution."],"attestations":[],"space":"model-runtimes"}