Imp.Optimizer.TrainingJobAdoption (Imp v0.5.0)

Adopts a verified, already-completed Imp.Clients.TrainingJob as a workflow step.

Adoption performs no training, trainer dispatch, fusion, or artifact mutation. It exists so an ordinary consumer can continue a composition such as BetterTogether from a durable weight artifact produced earlier. Construction binds the exact job, verified artifact, base program, and base-model identities. Execution verifies all four again before TrainingJob.rebind/3 starts the artifact's supported deployment.

Only providers with an Imp content verifier are accepted. Currently those are local MLX-LM fused artifacts and Imp↔TRL artifacts. A changed job, artifact tree, incoming program, or base model fails before rebinding.

Summary

Types

t()

Functions

new(job, program)

Binds a completed job to the exact compatible program that may later adopt it.

Types

t()

@type t() :: %Imp.Optimizer.TrainingJobAdoption{
  artifact_sha256: String.t(),
  base_model_identity: map(),
  job: Imp.Clients.TrainingJob.t(),
  job_sha256: String.t(),
  program_sha256: String.t()
}

Functions

new(job, program)

@spec new(Imp.Clients.TrainingJob.t(), struct()) :: t()

Binds a completed job to the exact compatible program that may later adopt it.

This constructor reads and verifies durable artifact contents but never starts a deployment or invokes a trainer.