To help users take advantage of rikai for their existing mlflow models, we want to make it easier for them to bring those into the catalog. In particular, we want to allow the user to specify the following default options for models they weren't able to use rikai.mlflow.<flavor>.log_model to record:
SET 'rikai.model.flavor'='pytorch'; // default model flavor
SET 'rikai.model.artifact_path'='model'; // default path relative to the run artifact uri
SET 'rikai.spec.version'='1.0'; // default rikai spec version (NOT the model version)
SET 'rikai.model.stage'='production'; // default model stage to filter for
CREATE MODEL my_model
OPTIONS (
'rikai.output.schema'='rikai.contrib.torch.transforms.fasterrcnn_resnet50_fpn.OUTPUT_SCHEMA',
'rikai.transforms.pre'='rikai.contrib.torch.transforms.fasterrcnn_resnet50_fpn.pre_processing',
'rikai.transforms.post'='rikai.contrib.torch.transforms.fasterrcnn_resnet50_fpn.post_processing'
)
USING 'mlflow://model_name'
To help users take advantage of rikai for their existing mlflow models, we want to make it easier for them to bring those into the catalog. In particular, we want to allow the user to specify the following default options for models they weren't able to use
rikai.mlflow.<flavor>.log_model
to record: