RIME v14 Changelog ================== AI Firewall ----------- * **Data Imputation** - The AI Firewall now automatically corrects model erroneous datapoints in real-time to prevent model downtime and reduce model performance deviations. * **Firewall Score** - Every datapoint that is passed to the AI Firewall is now given a Firewall Score that is a holistic measure of its quality. AI Continuous Testing --------------------- * **Key Insights** - The most significant underlying data errors that may be causing a degradation in model performance are now surfaced in the main Continuous Testing results page. Attribution of model behavior declines is now easier than ever. * **Production SDK** - Continuous Testing is ready for production deployment with full API support. This includes functionality to create CT endpoints, incrementally upload data, and query results. AI Stress Testing ----------------- * **Selectable Performance Metrics** - Users now have the flexibility to choose between performance metrics used in Tabular Abnormal Input and Subset Performance tests. * **Additional LTR Metric** - There is now a new performance metric (NDCG at K) in the LTR use case in addition to Mean Reciprocal Rank and Rank Correlation. The new NDCG at K metric is the new default metric used in the LTR use case is applied across Abnormal Input and Subset Performance tests. * **RIME Images** - RIME now supports the Image Classification use case. Reach out to the RI team to schedule a demo. * **Custom Tests** - Users can now create their own custom stress tests and run them on RIME. General ------- * **RIME Hosted Trial** \- Trials can now be conducted in a secure hosted environment that removes the requirement to download RIME. * **Performance Improvements** - There have been general improvements to RIME Performance that have led to a large increase in RIME production performance. * **RAM Usage Improvements** - RIME Engine has been optimized to use less memory to prevent OOM errors on data upload. * **Partitioned Dataset Upload** \- Users can now upload large partitioned datasets to RIME.