| """Substrate-neutral fine-tuning internals for the AutoSLM package. | |
| This subpackage holds the shared recipe, data loaders, graders, run accounting, | |
| and the on-GPU worker entrypoint. It has no dependency on any compute backend; the | |
| The RunPod provider in ``autoslm.providers.runpod`` invokes ``autoslm.engine.worker`` on the | |
| provisioned GPU. | |
| """ | |