Module handle 

Server – Handle

Runtime handle for a llama.cpp server process started by this crate.
The handle owns the child process and exposes a typed client so the rest of the crate can talk to the model without caring how it was launched.

Core ideas

• Process isolation – the server lives in a separate process, keeping crashes and memory leaks away from the host application.
• Deterministic boot – we distinguish loading, running and unhealthy/offline states and fail fast when the wrong model is seen.
• Time budgets – callers choose how patient to be for downloads, model loading and individual polling retries.

Dropping LmcppServer stops the external process automatically.

Structs

DownloadBudget

Maximum time the caller is willing to wait for a model download to complete when the server launches with a remote source (HF repo / URL).

LmcppServer

Handle representing a live llama.cpp server process.

LoadBudget

Maximum time the caller is willing to wait for the model to load after the server binary has started.

RetryDelay

Back‑off applied between repeated health probes during start‑up.

Enums

ServerStatus

Observable state of a llama.cpp server obtained via /health and /props.

Functions

model_ids_match

Decide whether two user‑supplied model identifiers refer to the same model file.