Chapter 6. Filtering and assessing models by tensor type

Procedure

1. In the OpenShift AI dashboard, click AI hub Models Catalog.
2. In the filter pane on the left, scroll down to the filter options for Tensor type.

3. Select one or more tensor types, for example, FP16 (16-bit floating point) and BF16 (brain floating point).

   The model catalog is filtered to show only models that have variants matching the selected tensor types.

   Note

   The Tensor type filter is always available and works regardless of whether the Model performance view toggle is enabled.

4. Click a validated model to display its model details.

5. Click the Performance Insights tab.

   This tab contains a Model variants by tensor type comparison card for compression variants based on tensor types.

6. Click a model compression variant to visually assess the trade-off between compression and quality for the selected model in the metrics displayed on the Performance Insights tab.

   Note

   If you enabled the Model performance view and set active filters in the catalog, those filters are automatically carried over to the Performance Insights tab.