Content

Target customers: customers who need to evaluate the model effect when deploying and fine-tuning models. New features/specifications: Provides a general-purpose model evaluation feature on the PAI platform, automatically assessing the overall capabilities of models. 1. Supports multiple evaluation targets: Public models from PAI-Model Gallery, custom models, PAI-EAS services, and custom services. 2. Evaluations can be performed using built-in authoritative public datasets (e.g., CMMLU, C-Eval, MMLU) or user-defined datasets. 3. Evaluation metrics include standard NLP metrics and referee model metrics. 4. Supports the comparison of multiple models. 5. Automatically generates evaluation reports.