This PR introduces a new configurable option for the Gemini Manifold pipeline, providing users with greater flexibility in managing content safety settings:
Added a new valve USE_PERMISSIVE_SAFETY (default: false) to the pipeline configuration.
When enabled, this setting applies the most permissive safety thresholds across all harm categories.
Users can now toggle between permissive and default safety settings without code modifications.
Key benefits:
Reduces unexpected errors for benign prompts that might trigger overly cautious filters.
Allows advanced users to handle content moderation at the application level if desired.
Maintains the option to use Google's default safety settings.
This change aims to balance user freedom with responsible AI use, giving Open-WebUI users more control over their interaction with the Gemini models while preserving the option for stricter content filtering when needed.
[!CAUTION]
Note: Users should be aware of and comply with Google's usage policies and any applicable regulations when adjusting these settings.
This PR introduces a new configurable option for the Gemini Manifold pipeline, providing users with greater flexibility in managing content safety settings:
USE_PERMISSIVE_SAFETY(default: false) to the pipeline configuration.Key benefits:
This change aims to balance user freedom with responsible AI use, giving Open-WebUI users more control over their interaction with the Gemini models while preserving the option for stricter content filtering when needed.
Addresses:
177
Credit:
@rotemdan for the code snippet in #177