llama.cpp has no paper - honest negative, verified against arXiv and OpenReview 2026-10-04

RECORDED AS A NEGATIVE because llama.cpp is one of the most consequential inference engines in the world and a reader of this repo will ask why there is no llama.cpp paper record: THERE ISN'T ONE, and inventing an arXiv ID for it would repeat the exact failure mode this repo has already shipped twice. Evidence 2026-10-04: export.arxiv.org/api/query search_query=all:"llama.cpp" returns 5 entries, none of which is a llama.cpp system paper - they are third-party papers that merely MENTION llama.cpp (2601.14277 quantization evaluation of llama.cpp, 2406.10816 Armv9 optimization study, 2511.05502 Apple Silicon comparative study, 2603.23996 forensic analysis of Ollama). No title search for 'llama.cpp' as a system returns anything. llama.cpp is a repository (github.com/ggml-org/llama.cpp), released March 2023, and has deliberately never published a paper. The repo therefore represents it through ENGINE and KERNEL source records rather than through data/papers/. This note exists so a future pass does not 'notice the gap' and fabricate a citation; the correct action on finding llama.cpp work worth recording is to add an engine or kernel source record, not a paper.