To be able to provision the models for the dane-asr-worker (in a k8s environment) it's best to isolate model download in a separate image, so we can do something like:
use the "model downloader" in a k8s job to download models in a dynamically provisioned pv
implement a ready/health check in the dane-asr-worker (if the models are not there, fail)
wire up the ready check to the dane-asr-worker image in the k8s manifest (deployment)
To be able to provision the models for the dane-asr-worker (in a k8s environment) it's best to isolate model download in a separate image, so we can do something like:
see also https://github.com/beeldengeluid/k8s-workloads/pull/34