You totally ignored the Lamini JSON output mode - which is full speed and supports enums for classifiers

https://lamini-ai.github.io/inference/json_output/