Skip to main content

Module runtime

clawft_kernel::workload_runtime::infer

Module runtime 

Source
Expand description

InferRuntime: the infer.llamacpp, infer.mlx-lm and infer.ollama adapters (ADR-101 section 4) behind one type.

Two constructions of the same adapter:

  • Adopted (InferConfig::adopted, id infer.<x>.adopted): the adapter registers and health-checks a server somebody else started. It never starts, stops or signals it: start only confirms the server answers, stop is refused, unload forgets the registration.
  • Managed (InferConfig::managed, id infer.<x>): llama.cpp and mlx-lm are launched through the model lab’s launcher under Supervised; Ollama is driven through its API. Weights come only from the model registry (crate::model_manifest::ModelRegistry), never from a path in the spec.

Adapters make no governance decisions; wrap one in crate::workload_runtime::WorkloadHost to gate and chain it.

Structs§

InferRuntime
The adapter.