Operational reference
LLM Systems
Models, artifacts, runtimes, and hardware — structured for people who self-host or deploy LLMs in practice.
What is catalogued
Open-weight model records, concrete GGUF artifacts (with publisher and size), the llama.cpp runtime, and hardware memory classes.
How claims are labeled
Every size and capability value carries an evidence class — SOURCE, POLICY, OBSERVED, and related — so measured facts stay distinct from site rules and unknowns.
What is not claimed yet
VRAM bands are weights-only from artifact file size. KV cache, runtime buffers, and full fit or max-context results are not modeled here.