Probing

Trains a logistic regression probe on frozen embeddings to evaluate what linguistic properties are encoded. Uses train/val/test splits defined per-sample in the dataset.

Metrics

• Per-task accuracy: accuracy on each individual probing task
• Average accuracy: mean accuracy across all probing tasks

CLI

steb probing rrivera1849/LUAR-MUD --dataset <dataset_name>

Episode size and episodes-per-class are set automatically for probing tasks.

Datasets

List available datasets:

steb probing --list-datasets