ml-training

Paths below are relative to this skill's directory.

Scope

Everything between "we have a model and some data" and "the weights we produced are serving traffic": method choice, the memory arithmetic that decides whether a run is possible, distributed strategy, dataset hygiene, evaluation that can be believed, checkpointing, and the handoff to an inference server.

Calling a model someone else trained is the ai-engineering skill's job; changing the weights is the ml-training skill's.

Not covered — do not answer from this skill: