cleanup, add example training package dist, changed container URI

Author: amygdala

ml/kubeflow-pipelines/keras_tuner/components/ucaip/model_deploy_component.yaml

@@ -10,7 +10,7 @@ inputs:
 - {name: timeout, type: Integer, default: '7200', optional: true}
 implementation:
   container:
-    image: gcr.io/aju-vtests2/bw-aiplatform:v1
+    image: gcr.io/google-samples/bw-aiplatform:v1
     command:
     - sh
     - -ec
@@ -87,7 +87,6 @@ implementation:
       #       project=project, location=location, endpoint=endpoint_id
       #   )
         response = client.deploy_model(
-          #   endpoint=endpoint, deployed_model=deployed_model, traffic_split=traffic_split
             endpoint=endpoint_path, deployed_model=deployed_model, traffic_split=traffic_split
         )
         logging.info("Long running operation: %s", response.operation.name)

ml/kubeflow-pipelines/keras_tuner/components/ucaip/model_train_component.yaml

@@ -19,7 +19,7 @@ outputs:
 - {name: model_dispname, type: String}
 implementation:
   container:
-    image: gcr.io/aju-vtests2/bw-aiplatform:v1
+    image: gcr.io/google-samples/bw-aiplatform:v1
     command:
     - sh
     - -ec
@@ -43,8 +43,6 @@ implementation:
           api_endpoint,  # "us-central1-aiplatform.googleapis.com",
           data_dir,
           hptune_dict,
-          # model_id: OutputPath('String'),
-          # model_dispname: OutputPath('String')
       ):
 
         import logging
@@ -97,6 +95,9 @@ implementation:
                     "replica_count": 1,
                     "container_spec": container_spec,
                 }
+        else:
+          logging.warning('unknown train_container_type; exiting')
+          exit(1)
 
         training_task_inputs_dict = {
             "workerPoolSpecs": [
@@ -110,8 +111,6 @@ implementation:
         training_task_inputs = json_format.ParseDict(training_task_inputs_dict, Value())
 
         training_task_definition = "gs://google-cloud-aiplatform/schema/trainingjob/definition/custom_task_1.0.0.yaml"
-        # image_uri = "gcr.io/cloud-aiplatform/prediction/tf-cpu.1-15:latest"
-        # image_uri = 'us-docker.pkg.dev/cloud-aiplatform/prediction/tf2-cpu.2-3:latest'
 
         training_pipeline = {
             "display_name": display_name,
@@ -139,21 +138,11 @@ implementation:
           logging.info('job state: %s', mresponse.state)
           if mresponse.state == pipeline_state.PipelineState.PIPELINE_STATE_FAILED:
             logging.warning('training pipeline failed: %s', mresponse)
-            break
+            exit(1)
           if mresponse.state == pipeline_state.PipelineState.PIPELINE_STATE_SUCCEEDED:
             logging.info('training finished')
             model_name = mresponse.model_to_upload.name
             return (model_name, model_display_name)
-            # # write some outputs once finished
-            # model_name = mresponse.model_to_upload.name
-            # logging.info('got model name: %s', model_name)
-            # with open('temp.txt', "w") as outfile:
-            #   outfile.write(model_name)
-            # subprocess.run(['gsutil', 'cp', 'temp.txt', model_id])
-            # with open('temp2.txt', "w") as outfile:
-            #   outfile.write(model_display_name)
-            # subprocess.run(['gsutil', 'cp', 'temp2.txt', model_dispname])
-            # break
           else:
             time.sleep(SLEEP_INTERVAL)
 

ml/kubeflow-pipelines/keras_tuner/components/ucaip/model_upload_component.yaml

@@ -11,7 +11,7 @@ outputs:
 - {name: model_id, type: String}
 implementation:
   container:
-    image: gcr.io/aju-vtests2/bw-aiplatform:v1
+    image: gcr.io/google-samples/bw-aiplatform:v1
     command:
     - sh
     - -ec
@@ -28,7 +28,6 @@ implementation:
           location, # "us-central1",
           api_endpoint, #"us-central1-aiplatform.googleapis.com",
           timeout, # 1800,
-          # model_id: OutputPath('String')
           ):
 
         import logging

ml/kubeflow-pipelines/keras_tuner/components/ucaip/serving/deploy_model.py

@@ -80,18 +80,16 @@ def create_endpoint(
 #       project=project, location=location, endpoint=endpoint_id
 #   )
   response = client.deploy_model(
-    #   endpoint=endpoint, deployed_model=deployed_model, traffic_split=traffic_split
       endpoint=endpoint_path, deployed_model=deployed_model, traffic_split=traffic_split
   )
   logging.info("Long running operation: %s", response.operation.name)
   deploy_model_response = response.result(timeout=timeout)
   logging.info("deploy_model_response: %s", deploy_model_response)
+  # TODO: output status info in some form
 
 
 if __name__ == '__main__':
-#   deploy_model('aju-vtests2', 'endpoint_test2',
-#       'projects/467744782358/locations/us-central1/models/6181278449496227840', 'sdk_test1')
   import kfp
   kfp.components.func_to_container_op(deploy_model,
       output_component_file='../model_deploy_component.yaml',
-      base_image='gcr.io/aju-vtests2/bw-aiplatform:v1')
+      base_image='gcr.io/google-samples/bw-aiplatform:v1')

ml/kubeflow-pipelines/keras_tuner/components/ucaip/serving/model_upload.py

@@ -24,7 +24,6 @@ def upload_model(
     location: str, # "us-central1",
     api_endpoint: str, #"us-central1-aiplatform.googleapis.com",
     timeout: int, # 1800,
-    # model_id: OutputPath('String')
     ) -> NamedTuple('Outputs', [('model_id', str)]):
 
   import logging
@@ -64,21 +63,14 @@ def upload_model(
   logging.info("upload_model_response: %s", upload_model_response)
   model_path = upload_model_response.model
   return (model_path, )
-  # logging.info('got model path: %s', model_path)
-  # with open('temp.txt', "w") as outfile:
-  #   outfile.write(model_path)
-  # subprocess.run(['gsutil', 'cp', 'temp.txt', model_id])
 
 
 
 if __name__ == '__main__':
-  # upload_model('aju-vtests2', display_name='sdk_test1', metadata_schema_uri="",
-  #     image_uri='us-docker.pkg.dev/cloud-aiplatform/prediction/tf2-cpu.2-3:latest',
-  #     artifact_uri='gs://aju-pipelines/v64/077ae97e-9c6d-4c1c-b5a1-fc2e95fb7dbb/0/bwmodel/trained_model/export/bikesw/1615937808')
   import kfp
   kfp.components.func_to_container_op(upload_model,
       output_component_file='../model_upload_component.yaml',
-      base_image='gcr.io/aju-vtests2/bw-aiplatform:v1')
+      base_image='gcr.io/google-samples/bw-aiplatform:v1')
 
 
 # gcloud beta ai models upload --region=us-central1 --display-name=bw2 --container-image-uri=us-docker.pkg.dev/cloud-aiplatform/prediction/tf2-cpu.2-3:latest --artifact-uri=gs://aju-pipelines/ktune13/f8515c75-32b7-47a4-af70-5ff24362eccc/0/bwmodel/trained_model/export/bikesw/1603733739

ml/kubeflow-pipelines/keras_tuner/components/ucaip/training/create_training_job.py

@@ -32,8 +32,6 @@ def create_training_pipeline_custom_job(
     api_endpoint: str,  # "us-central1-aiplatform.googleapis.com",
     data_dir: str,
     hptune_dict: str,
-    # model_id: OutputPath('String'),
-    # model_dispname: OutputPath('String')
 ) -> NamedTuple('Outputs', [('model_id', str), ('model_dispname', str)]):
 
   import logging
@@ -102,8 +100,6 @@ def create_training_pipeline_custom_job(
   training_task_inputs = json_format.ParseDict(training_task_inputs_dict, Value())
 
   training_task_definition = "gs://google-cloud-aiplatform/schema/trainingjob/definition/custom_task_1.0.0.yaml"
-  # image_uri = "gcr.io/cloud-aiplatform/prediction/tf-cpu.1-15:latest"
-  # image_uri = 'us-docker.pkg.dev/cloud-aiplatform/prediction/tf2-cpu.2-3:latest'
 
   training_pipeline = {
       "display_name": display_name,
@@ -136,31 +132,11 @@ def create_training_pipeline_custom_job(
       logging.info('training finished')
       model_name = mresponse.model_to_upload.name
       return (model_name, model_display_name)
-      # # write some outputs once finished
-      # model_name = mresponse.model_to_upload.name
-      # logging.info('got model name: %s', model_name)
-      # with open('temp.txt', "w") as outfile:
-      #   outfile.write(model_name)
-      # subprocess.run(['gsutil', 'cp', 'temp.txt', model_id])
-      # with open('temp2.txt', "w") as outfile:
-      #   outfile.write(model_display_name)
-      # subprocess.run(['gsutil', 'cp', 'temp2.txt', model_dispname])
-      # break
     else:
       time.sleep(SLEEP_INTERVAL)
 
 if __name__ == '__main__':
-  # create_training_pipeline_custom_job(
-  #   'aju-vtests2', 'bw_sdktest2',
-  #   'bw_sdktest2',
-  #   'us-docker.pkg.dev/cloud-aiplatform/training/tf-gpu.2-3:latest',
-  #   'gs://aju-pipelines/ucaip/training1/bw-trainer-0.1.tar.gz',
-  #   'trainer.task',
-  #   'gs://aju-pipelines/ucaip/test1803_sdk1',
-  #   "us-central1",
-  #   "us-central1-aiplatform.googleapis.com",
-  # )
   import kfp
   kfp.components.func_to_container_op(create_training_pipeline_custom_job,
       output_component_file='../model_train_component.yaml',
-      base_image='gcr.io/aju-vtests2/bw-aiplatform:v1')
+      base_image='gcr.io/google-samples/bw-aiplatform:v1')