Skip to main content

Exporting a model

Exporting packages a trained LoRA adapter into a downloadable archive so you can keep a copy outside EcoLink — to archive it, run it in your own environment, or move it between accounts. Deploying an adapter (attaching it to a model instance) is separate; see Deploying an adapter.

Export an adapter​

  1. Go to /fine-tuned-models (or open the fine-tune job's detail page).
  2. Click Export on the adapter you want.
  3. EcoLink starts an in-region export job that packages the adapter. This runs asynchronously — the status updates as it works.
  4. When it's ready, a download link appears. The archive is a .tar.gz containing the LoRA adapter weights and config.

The download link is time-limited — download the file promptly. You can re-export the same adapter later to get a fresh link.

What's in the archive​

The archive holds the adapter itself — the LoRA weights and their configuration — not the base model. To use the adapter you pair it with the same base model it was trained on, either back on EcoLink (by deploying it) or in your own LoRA-capable serving stack.

Exporting files from storage​

You can also export a folder from a cloud drive or shared filesystem — useful for pulling out datasets, checkpoints, or other artifacts a GPU job produced. Pick a specific folder (not the whole drive), start the export, and download the resulting .tar.gz when it completes. The flow mirrors adapter export: start → the job packages the folder → a download link appears.