Feature: Support Google LLM safety_settings param #3732
Description
Feature Type
Would make my life easier
Feature Description
Gemini supports various content filtering parameters under safety_settings in the GenerateContentConfig call to restrict the LLM output based on different HarmCategory definitions and HarmBlockThreshold levels.
Could we update the google llm.py plugin so we can pass through the safety_settings as a param in the google.LLM() constructor?
Workarounds / Alternatives
- Filtering and restrictions via prompts
- Secondary agent watching text stream to monitor for safety
Additional Context
- https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/inference#harmcategory
- https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/configure-safety-filters#gemini-api-in-vertex-ai