ml-commons
ml-commons copied to clipboard
[FEATURE] Add Recommended Space Type to OpenSearch-provided Pretrained Dense Models
Is your feature request related to a problem? When use KNN query with a dense model it's not easy to decide what space type we should use. If the recommended space type can be part of the model metadata for OpenSearch-provided Pretrained Dense Models it can help users who are not very familiar with ML models and KNN query.
We have an issue in KNN plugin about this. And we also have a feature request in neural search plugin that expects us to be able to automatically decide the space type for the dense models.
What solution would you like? For the OpenSearch-provided Pretrained Dense Models we should be able to know the recommended space type for KNN query from the model metadata.
For custom model we should allow users provide the recommended space type. Let's say a model owner want to recommend the space type there should be a way to add this info and then other people who want to use that model can easily access that info.
What alternatives have you considered? We can document/hardcode the recommended space type.
Do you have any additional context? N/A