Examples for deployment of ML models on K8s via Seldon Core. Order according from simple to sohpisticated.
-
01_getting-started
: deploys iris model binary provided by Seldon on Google Cloud Storage with pre-packaged seldon sklearn MLServer.
-
02_getting-started-custom
: Trains iris model, use custom language wrapper (predict class) and build image with s2i, use with MLServer.
-
03_getting-started-custom2
: as example2_getting-started-custom
but build image with regular docker file instead of s2i.
-
04_getting-started-custom-mlflow/
: Trains iris model and saves to mlflow. Loads separate Mlflow Cloud Storage model as deploys with pre-packaged seldon mlflow MLServer.
-
05_inference-graph-randab
: Demos random A/B test with two models leveraging the seldon RANDOM_ABTEST implementation.
-
06_canary-rollout
: Demos the application of two models via a Canary deployment.
-
07_seldon-grafana-metrics
: Simple deployment with notes about working with Prometheus/Grafana.
-
08_sklearnserver
: effort to build custom seldon inference server. Not pursued further as implementation requires modification of seldon core operator settings, which is unpractical for deployment at this point.
-
09_text_classifier_mlflow
: trainings NLP classifier model with some dependencies and saves as custom Python mlflow model. then deployment via pre-packaged seldon mlflow server.
-
10_translator_mlflow
: deploys Transformers French-to-English language translator model. Model mlflow artifact is pulled from minio on k8s.