vram

Manage local LLM inference servers in DSH: GPU registry, parameter profiles, VRAM validation, and tok/s benchmarks for llama.cpp, SGLang, and vLLM