llm-hosting

name	llm-hosting
description	Use for llama.cpp, llama-server, local model hosting, ROCm or Vulkan inference, model and KV-cache tuning, Open WebUI, or OpenCode local-provider configuration in this repository.

LLM Hosting

Read the relevant section of docs/LLM-HOSTING-TUNING.md; search its headings first instead of loading unrelated sections.

For OpenCode and nixvim integration, also read docs/NIXVIM.md and inspect modules/home/nixvim/default.nix. For NixOS service behavior, inspect modules/nixos/llama-cpp/default.nix.

Preserve measured hardware-specific values unless the task provides new evidence. After edits, run ./format.fish; never run a Nix build or activation command.