pub const COMPLETION: &str = "/completion";Expand description
llama.cpp’s native completion endpoint, which is not
V1_COMPLETIONS with a shorter path.
Different request shape (n_predict, repeat_penalty,
cache_prompt, …) and a different response shape (a flat object
with content and stop, not choices), and its stream is a
sequence of data: {"content":…,"stop":false} frames with no
[DONE] sentinel. This is what llama.cpp’s own web UI, llama.vim
and a long tail of wrappers speak; /v1/completions is the OpenAI
dialect and stays what it is.