Utility Scripts

The ground segment includes two shell scripts for managing data and model weights across local and remote training servers.

upload_to_server.sh

ground_segment/data/upload_to_server.sh uploads the generated dataset to a remote GPU server for training.

What it does:

1. Compresses orion_dataset/ into a tarball
2. Transfers it to the server's HDD via rsync
3. Unpacks on the server, removing any previous dataset
4. Clones or pulls the ORION repository on the server

Requires: The SERVER environment variable set to the remote server address (e.g., export SERVER=user@gpu-server).

Server paths (hardcoded):

Path	Purpose
~/hdd/gaze/datasets/extras/	Dataset storage on server HDD
~/code/extras/ORION/	Repository clone on server

download_weights.sh

ground_segment/training/download_weights.sh downloads trained LoRA weights from the remote server after fine-tuning completes.

What it does:

1. Verifies the weights directory exists on the server
2. Wipes any existing local weights to prevent stale data
3. Downloads orion_lora_weights/ via rsync
4. On success, cleans up the repository and dataset on the server (scorched earth)

Requires: The SERVER environment variable set to the remote server address.

Important: On successful download, this script deletes the repository and dataset from the server. This is intentional — the weights are the only artifact needed after training.