Reputation: 6289
--resolved -- by using cli to UNDEPLOY the model then following docs https://cloud.google.com/vertex-ai/docs/image-data/classification/get-predictions#curl
to submit batch-prediction
vertex-ai newbie just trying to learn basics without alot of cost, using the vertex-ai dashboard to good effect ( training model and endpoint to recognize/ label photos using 3 categs from training labels)
but the problem i experience was that somehow , inadvertently thru the dashboard, the model was deployed and led to daily $30 charges to the project.
IMO - from the vertex-ai docs, you should be able to run batch-predictions WITHOUT having to deploy the model/ endpoint.
Is anyone able to verify the above?
with the current price lists, on a very small budget its possible to touch most of the bases ( create, train, test a model to label photos ) as long as you DO NOT deploy it. AND batch predicitions appear to be allowed with NO deployment.
The only costly SKU i ran into was the actual deployment see photo
Upvotes: 1
Views: 104
Reputation: 1377
--resolved -- by using cli to UNDEPLOY the model then following docs https://cloud.google.com/vertex-ai/docs/image-data/classification/get-predictions#curl to submit batch-prediction
For the undeploy command details you can visit this documentation here.
Sample command:
gcloud ai endpoints undeploy-model 123 --project=example --region=us-central1 --deployed-model-id=456
Upvotes: 0