Sometimes the solution training happens to be successful but the prediction is failing due to the artifact size exceeded the limit. A trainer has a max allowed limit of 40MB but the prediction has a 20MB limit. This is applicable for pre-New York releases only.
Steps to Reproduce
- Log into the instance and go to REST API Explorer;
- Leave Namespace as "now", select "Agent Intelligence" as API Name and leave API Version as "latest";
- In the "solution_name" field, input the name of the Solution trained
- Populate the Query Parameters as needed
- Click on Send
Expected behavior: Output of the JSON state the outcome, confidence and threshold
Current behavior: Output of the JSON state the outcome as "Unable to predict"
In the logs we can see an error that states:
Prediction API call failed, possible reason is artifact size is greater than the value specified in glide property 'glide.platform_ml.artifact.attachment_size'.
Please increase it to a value greater than the current artifact size or contact support.
Create or update the System Property [ sys_properties ] "glide.platform_ml.artifact.attachment_size" and set the Value to 40MB. The value in this System Property is in Bytes, so 40MB should be inserted as 41943040.
Related Problem: PRB1347994