Expand description
Composite engine that delegates per task kind.
Operators set engine = "multi" and engines = [...] in their
config to combine multiple inference backends in a single binary:
engine = "multi"
engines = ["llama", "tts", "video", "synthetic"]For each task kind, MultiEngine picks the first engine in the
list that advertises support for the requested model. If no engine
claims it, the dispatch fails with the same “cannot serve