Early Stopping in Prompt Tuning

[alex-jw-brooks]

Background

Currently, we have no early stopping added to the prompt tuning module, but we should! This issue covers adding it.

Acceptance Criteria

• [ ] Prompt tuning can be run with early stopping
• [ ] Tests added on tiny models validating early stopping via stubbed behaviors or with (fast) deterministic examples

[chinghuachen]

This is related to https://github.ibm.com/ai-foundation/watson-fm-stack-tracker/issues/302

[chinghuachen]

WCA for Red Hat Ansible Lightspeed has a need for this feature, with the clarification that the tests are performed on a validation test set that is distinct from the training data.