Add safety checking infrastructure for text generation #41776
+5,042
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Provides infrastructure for runtime safety checking via safety_config parameter. Includes base classes, configuration, and processors. Users implement concrete checkers for their specific needs.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Draft PR: Core implementation complete. Seeking feedback on design and approach before finalizing. Thank you!
What does this PR do?
Adds safety checking infrastructure for text generation. Provides base classes, configuration, and processors that integrate with the generation pipeline. Users implement their own safety checkers for specific needs (harm, bias, PII, etc.).
Fixes #41740
Motivation
As stated in the issue I opened, while production LLMs have built-in safety moderation systems, they are often insufficient and can lead to unexpected harmful behavior, especially over long conversations. As open-source text generation models become more capable and widely used, mitigating harm and ensuring user safety is a feature that should be built in. As far as I am aware, there is currently no built-in infrastructure to support this. The most effective approaches involve moderation during inference, which is a non-trivial feature for Transformers users to implement on their own. In addition, allowing for the configuration of safety with custom settings and classifiers can allow users to avoid harm in more specialized contexts than commercial LLMs currently address.
Before submitting
Pull Request section?
to it if that's the case. Currently being discussed in Safety Checking Infrastructure for Text Generation #41740
documentation guidelines, and
here are tips on formatting docstrings.