For Multimodel Serving
Troubleshoot issues when using multimodel serving with model deployments.
Model loading fails due to invalid models
In the predict logs is the error: Failed to load model
.
A faulty model exists in the model group.
- Fix the faulty model or remove it from the model group.
- Deploy the updated model group.
Live update request can't be processed
The model group size is larger than the maximum supported size.
Use a ZDT deployment instead.
Live update can't be fulfilled because of a change in the model group type
Live Update isn't supported when switching between different Model Group types, for example, from HOMOGENEOUS
to STACKED
.
Perform a ZDT update.
Deployment type can't be changed during an update
The deployment type can't be changed for an update operation.
Create a new model deployment with the required
deploymentType
(SINGLE_MODEL
or MODEL_GROUP
).
Live update fails if the update request isn't for modelGroupId
Live update is supported for updating modelGroupId
only.
Perform the live update for updating
modelGroupId
only. A new ModelGroupId
must have differences in member models
only. The rest of the configuration must remain same as the deployed model group.
Model group creation fails because of invalid models
The models don't exist or are in a terminal state.
Use existing models not in a terminal state.