Oracle Guardian AI 1.3.0

Added new support for LLMs in the fairness module for measuring toxicity bias in LLMs. These metrics measure the disparity in toxic generations -- that is, whether or not your LLM is more toxic when talking about one group of people than another.

Oracle Guardian AI 1.2.0

• Upgraded scikit-learn to 1.5.0

Oracle Guardian AI 1.1.0

• Enhanced bias mitigation to avoid solutions with levelling down (that is, making outcomes worse for) one or more groups to achieve fairness metric rate parity.
• Added warm starting mechanism to bias mitigation to reduce the time required to find high-quality solution trade-offs.
• Replaced AIF360 rate-based fairness metrics with in-house ones to improve running times.

Oracle Guardian AI 1.0.1

• Fixed a bug in the rate-based fairness metrics that caused them to report incomplete results when using reduction=None.

Oracle Guardian AI 1.0.0

Oracle Guardian AI Open Source Project is a library consisting of tools to assess fairness/bias and privacy of machine learning models and data sets.