Off-Topic Classification/Evaluation and usage of all-MiniLM-L6-v2

[smrazaabbas]

Hi,

I'm trying to implement NemoGuardrails and trying to work with the nemoguardrails/eval/evaluate_topical.py ; are these utilities in the eval folder only for the evaluation of GuardRails App or do they serve any other purpose?

What I'm really interested in is trying to see if there's a possibility of using self._model = SentenceTransformer("all-MiniLM-L6-v2") and related vector search utilities to find off-topic prompts. Is there a way to turn off OpenAI and just stick to sentence transformers for the Off-topic classification.