| """RunPod Flash integration: managed, serverless GPUs (no Docker) for AutoSLM. | |
| Fine-tuning runs on a dedicated RunPod GPU (RTX 4090 / RTX 5090) provisioned by Flash. | |
| A decorated Python handler (``autoslm.flash.train``) executes the ``autoslm.engine.worker`` | |
| on the GPU; Flash handles provisioning, dependency install, execution, and scale-to-zero | |
| teardown. Serving (``autoslm.flash.serve``) exposes an OpenAI-compatible endpoint for a | |
| trained LoRA adapter. | |
| """ | |