FEAT: Adding Harm Categories to Prompt Request Pieces

[jbolor21]

Description

Making it so we can query attack_results by harm_categories and memory labels. This value is currently present in seed prompts but was not queryable for attack results. To do this I made a few changes:

• Adding Harm_Categories to PromptRequestPieces
• Adding harm_categories into get_attack_results() so we can query by harm_categories for attacks. This is built on logic joining two datatables together
• Adding labels into get_attack_results() similar to above
• A notebook to demonstrate how to query by harm_category, and added this into the prompt sending cookbook as well
• Updated illegal.yaml to change the money laundering prompt instead to a violent prompt so we can easily demonstrate our queries for both single and multiple harm categories.

Tests and Documentation

Ran notebooks, added new unit tests

[romanlutz]

This is a good start! I think we should also have an example showing how to query by harm category within a specific op label, and the memory code needs a join between prompt memory entry and attack results to check for all results with a certain harm category in the pieces.

[hannahwestra25]

nice work! made a few small comments but overall looks good!

[hannahwestra25]

two small comments on the comments :)