Model Gallery provides an entry point for evaluating large language models. You can evaluate model capabilities by using custom datasets or public datasets.
Feature scope
Model Gallery supports only single model evaluation, which scores the online inference results of a single model. For dual model battles (Dual-Model Arena and Dual-Model Offline Competition), go to Model evaluation (ModelEval).
Evaluation tasks created through Model Gallery can be viewed and managed in either of the following locations:
Model Gallery > Job Management > Evaluation Jobs
- Model Evaluation (ModelEval)
Create an evaluation task
Log on to the PAI console. In the left-side navigation pane, choose QuickStart > Model Gallery.
-
In the Models page, select Evaluate from the Supported Operations filter to show only evaluable models.
NoteModels derived from fine-tuning a model that supports evaluation also support evaluation. In Job Management > Training Jobs, click a task to go to its details page. The Evaluate button is available in the upper-right corner.
Click the target model card to go to its details page, and then click Evaluate in the upper-right corner of the page.
For a full description of evaluation modes and parameters, see Model evaluation (ModelEval).