google-cloud-pipeline-components 2.14.0__py3-none-any.whl → 2.15.0__py3-none-any.whl

google_cloud_pipeline_components/_implementation/starry_net/upload_model/component.py

@@ -0,0 +1,23 @@
+# Copyright 2024 The Kubeflow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Starry Net Upload Model Component."""
+
+import os
+
+from kfp import components
+
+# TODO(b/346580764)
+upload_model = components.load_component_from_file(
+    os.path.join(os.path.dirname(__file__), 'upload_model.yaml')
+)

google_cloud_pipeline_components/_implementation/starry_net/upload_model/upload_model.yaml

@@ -0,0 +1,37 @@
+name: model_upload
+inputs:
+- {name: project, type: String}
+- {name: location, type: String, default: "us-central1"}
+- {name: display_name, type: String}
+- {name: description, type: String, optional: true, default: ''}
+- {name: unmanaged_container_model, type: google.UnmanagedContainerModel, optional: true}
+- {name: encryption_spec_key_name, type: String, optional: true, default: ''}
+- {name: labels, type: JsonObject, optional: true, default: '{}'}
+- {name: parent_model, type: google.VertexModel, optional: true}
+outputs:
+- {name: model, type: google.VertexModel}
+- {name: gcp_resources, type: String}
+implementation:
+  container:
+    image: gcr.io/ml-pipeline/automl-tables-private:1.0.17
+    command: [python3, -u, -m, launcher]
+    args: [
+      --type, UploadModel,
+      --payload,
+      concat: [
+          '{',
+          '"display_name": "', {inputValue: display_name}, '"',
+          ', "description": "', {inputValue: description}, '"',
+          ', "encryption_spec": {"kms_key_name":"', {inputValue: encryption_spec_key_name}, '"}',
+          ', "labels": ', {inputValue: labels},
+          '}'
+      ],
+      --project, {inputValue: project},
+      --location, {inputValue: location},
+      --gcp_resources, {outputPath: gcp_resources},
+      --executor_input, "{{$}}",
+      {if: {
+        cond: {isPresent: parent_model},
+        then: ["--parent_model_name", "{{$.inputs.artifacts['parent_model'].metadata['resourceName']}}",]
+      }},
+    ]

google_cloud_pipeline_components/_implementation/starry_net/version.py

@@ -0,0 +1,18 @@
+# Copyright 2024 The Kubeflow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Version constants for starry net components."""
+
+DATAPREP_VERSION = '20240617_2225_RC00'
+PREDICTOR_VERSION = '20240617_2142_RC00'
+TRAINER_VERSION = '20240617_2142_RC00'

google_cloud_pipeline_components/container/utils/error_surfacing.py

@@ -0,0 +1,45 @@
+# Copyright 2024 The Kubeflow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Utilities for surface user defined error messages."""
+
+import json
+import os
+from google.protobuf import json_format
+from google_cloud_pipeline_components.proto import task_error_pb2
+
+
+def write_customized_error(
+    executor_input: str, error: task_error_pb2.TaskError
+):
+  """Writes a TaskError customized by the author of the pipelines to a JSON file ('executor_error.json') in the output directory specified in the executor input.
+
+  Args:
+    executor_input: JSON string containing executor input data.
+    error: TaskError protocol buffer message.
+  """
+  executor_input_json = json.loads(executor_input)
+  os.makedirs(
+      os.path.dirname(executor_input_json['outputs']['outputFile']),
+      exist_ok=True,
+  )
+  executor_out_path = executor_input_json['outputs']['outputFile']
+  directory_path = os.path.dirname(executor_out_path)
+  executor_error_path = os.path.join(directory_path, 'executor_error.json')
+  error_dict = json_format.MessageToDict(error)
+  with open(
+      executor_error_path,
+      'w',
+  ) as f:
+    json.dump(error_dict, f)

google_cloud_pipeline_components/container/v1/model/get_model/remote_runner.py

@@ -12,13 +12,33 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Remote runner for Get Model based on the Vertex AI SDK."""
-
+import contextlib
+from typing import Tuple, Type, Union
 from google.api_core.client_options import ClientOptions
 from google.cloud import aiplatform_v1 as aip_v1
 from google_cloud_pipeline_components.container.utils import artifact_utils
+from google_cloud_pipeline_components.container.utils import error_surfacing
+from google_cloud_pipeline_components.proto import task_error_pb2
 from google_cloud_pipeline_components.types import artifact_types
 
 
+@contextlib.contextmanager
+def catch_write_and_raise(
+    executor_input: str,
+    exception_types: Union[
+        Type[Exception], Tuple[Type[Exception], ...]
+    ] = Exception,
+):
+  """Context manager to catch specified exceptions, log them using error_surfacing, and then re-raise."""
+  try:
+    yield
+  except exception_types as e:
+    task_error = task_error_pb2.TaskError()
+    task_error.error_message = str(e)
+    error_surfacing.write_customized_error(executor_input, task_error)
+    raise
+
+
 def get_model(
     executor_input,
     model_name: str,
@@ -26,11 +46,16 @@ def get_model(
     location: str,
 ) -> None:
   """Get model."""
-  if not location or not project:
-    raise ValueError(
-        'Model resource name must be in the format'
-        ' projects/{project}/locations/{location}/models/{model_name}'
-    )
+  with catch_write_and_raise(
+      executor_input=executor_input,
+      exception_types=ValueError,
+  ):
+    if not location or not project:
+      model_name_error_message = (
+          'Model resource name must be in the format'
+          ' projects/{project}/locations/{location}/models/{model_name}'
+      )
+      raise ValueError(model_name_error_message)
   api_endpoint = location + '-aiplatform.googleapis.com'
   vertex_uri_prefix = f'https://{api_endpoint}/v1/'
   model_resource_name = (
@@ -40,7 +65,11 @@ def get_model(
   client_options = ClientOptions(api_endpoint=api_endpoint)
   client = aip_v1.ModelServiceClient(client_options=client_options)
   request = aip_v1.GetModelRequest(name=model_resource_name)
-  get_model_response = client.get_model(request)
+  with catch_write_and_raise(
+      executor_input=executor_input,
+      exception_types=Exception,
+  ):
+    get_model_response = client.get_model(request)
   resp_model_name_without_version = get_model_response.name.split('@', 1)[0]
   model_resource_name = (
       f'{resp_model_name_without_version}@{get_model_response.version_id}'

google_cloud_pipeline_components/preview/llm/infer/component.py

@@ -18,7 +18,7 @@ from typing import NamedTuple, Optional
 from google_cloud_pipeline_components import _placeholders
 from google_cloud_pipeline_components._implementation.llm import bulk_inferrer
 from google_cloud_pipeline_components._implementation.llm import env
-from google_cloud_pipeline_components._implementation.llm import function_based
+from google_cloud_pipeline_components._implementation.llm import infer_preprocessor
 from google_cloud_pipeline_components._implementation.llm import preprocess_chat_dataset
 from google_cloud_pipeline_components._implementation.llm import private_text_importer
 import kfp
@@ -66,14 +66,16 @@ def infer_pipeline(
   """
   # fmt: on
   prompt_column = 'input_text'
-  machine_spec = function_based.resolve_machine_spec(
+  preprocess_metadata = infer_preprocessor.infer_preprocessor(
+      large_model_reference=large_model_reference,
       accelerator_type=accelerator_type,
       use_test_spec=env.get_use_test_machine_spec(),
-  ).set_display_name('Resolve Machine Spec')
-  reference_model_metadata = function_based.resolve_reference_model_metadata(
-      large_model_reference=large_model_reference,
-      reference_model_path=model_checkpoint,
-  ).set_display_name('Resolve Model Metadata')
+      project=env.PRIVATE_ARTIFACT_REGISTRY_PROJECT,
+      location=env.PRIVATE_ARTIFACT_REGISTRY_LOCATION,
+      artifact_registry=env.PRIVATE_ARTIFACT_REGISTRY,
+      tag=env.get_private_image_tag(),
+      instruction=instruction,
+  ).set_display_name('Preprocess Inputs')
 
   processed_dataset = preprocess_chat_dataset.preprocess_chat_dataset(
       large_model_reference=large_model_reference,
@@ -82,10 +84,6 @@ def infer_pipeline(
       dataset_type='prompt',
   ).set_display_name('Preprocess Dataset')
 
-  resolved_text_instruction = function_based.resolve_instruction(
-      large_model_reference=large_model_reference,
-      instruction=instruction,
-  ).set_display_name('Resolve Instruction')
   prompt_dataset_importer = (
       private_text_importer.private_text_importer(
           project=project,
@@ -94,35 +92,34 @@ def infer_pipeline(
           inputs_field_name=prompt_column,
           targets_field_name='',  # ignore targets_field_name
           output_split_name=env.TRAIN_SPLIT,
-          large_model_reference=reference_model_metadata.outputs[
-              'large_model_reference'
+          large_model_reference=preprocess_metadata.outputs[
+              'metadata_large_model_reference'
           ],
-          instruction=resolved_text_instruction.output,
+          instruction=preprocess_metadata.outputs['metadata_instruction'],
           encryption_spec_key_name=encryption_spec_key_name,
       )
       .set_display_name('Import Prompt Dataset')
       .set_caching_options(False)
   )
 
-  bulk_inferrer_image_uri = function_based.resolve_private_refined_image_uri(
-      accelerator_type=machine_spec.outputs['accelerator_type'],
-  ).set_display_name('Resolve Bulk Inferrer Image URI')
   bulk_inference = bulk_inferrer.bulk_inferrer(
       project=project,
-      location=machine_spec.outputs['tuning_location'],
-      input_model=reference_model_metadata.outputs['reference_model_path'],
+      location=preprocess_metadata.outputs['metadata_tuning_location'],
+      input_model=preprocess_metadata.outputs['metadata_reference_model_path'],
       input_dataset_path=prompt_dataset_importer.outputs['imported_data_path'],
       dataset_split=env.TRAIN_SPLIT,
       inputs_sequence_length=prompt_sequence_length,
       targets_sequence_length=target_sequence_length,
-      large_model_reference=reference_model_metadata.outputs[
-          'large_model_reference'
+      large_model_reference=preprocess_metadata.outputs[
+          'metadata_large_model_reference'
       ],
       sampling_strategy=sampling_strategy,
-      accelerator_type=machine_spec.outputs['accelerator_type'],
-      accelerator_count=machine_spec.outputs['accelerator_count'],
-      machine_type=machine_spec.outputs['machine_type'],
-      image_uri=bulk_inferrer_image_uri.output,
+      accelerator_type=preprocess_metadata.outputs['metadata_accelerator_type'],
+      accelerator_count=preprocess_metadata.outputs[
+          'metadata_accelerator_count'
+      ],
+      machine_type=preprocess_metadata.outputs['metadata_machine_type'],
+      image_uri=preprocess_metadata.outputs['metadata_refined_image_uri'],
       encryption_spec_key_name=encryption_spec_key_name,
   ).set_display_name('Bulk Inferrer')
 

google_cloud_pipeline_components/preview/llm/rlhf/component.py

@@ -106,6 +106,9 @@ def rlhf_pipeline(
       tag=env.get_private_image_tag(),
       evaluation_dataset=eval_dataset,
       tensorboard_resource_id=tensorboard_resource_id,
+      upload_location=location,
+      model_display_name=model_display_name,
+      deploy_model=deploy_model,
   ).set_display_name('Preprocess Inputs')
   num_microbatches = preprocess_metadata.outputs['metadata_num_microbatches']
 
@@ -133,6 +136,9 @@ def rlhf_pipeline(
               reward_model_image_uri=preprocess_metadata.outputs[
                   'metadata_refined_image_uri'
               ],
+              comma_separated_candidates_field_names=preprocess_metadata.outputs[
+                  'metadata_candidate_columns_string'
+              ],
               prompt_sequence_length=prompt_sequence_length,
               target_sequence_length=target_sequence_length,
               eval_dataset=validate_pipeline_task.outputs[
@@ -200,12 +206,9 @@ def rlhf_pipeline(
       has_inference_dataset == True,  # pylint: disable=singleton-comparison
       name='Perform Inference',
   ):
-    has_model_checkpoint = function_based.value_exists(
-        value=rl_model_pipeline.outputs['output_model_path']
-    ).set_display_name('Resolve Model Checkpoint')
-    with kfp.dsl.Condition(
-        has_model_checkpoint.output == True,  # pylint: disable=singleton-comparison
-        name='Test Model Checkpoint Exists',
+    with kfp.dsl.If(
+        rl_model_pipeline.outputs['output_model_path'] != '',
+        name='CheckModel Checkpoint Exists',
     ):
       component.infer_pipeline(
           project=project,
@@ -226,10 +229,14 @@ def rlhf_pipeline(
       policy_model_reference=preprocess_metadata.outputs[
           'metadata_large_model_reference'
       ],
-      model_display_name=model_display_name,
-      deploy_model=deploy_model,
+      model_display_name=preprocess_metadata.outputs[
+          'metadata_model_display_name'
+      ],
+      deploy_model=preprocess_metadata.outputs['metadata_deploy_model'],
+      upload_model=preprocess_metadata.outputs['metadata_upload_model'],
       encryption_spec_key_name=encryption_spec_key_name,
       upload_location=location,
+      regional_endpoint=preprocess_metadata.outputs['metadata_upload_location'],
   ).set_display_name('Upload and Deploy Tuned Model')
 
   return PipelineOutput(

google_cloud_pipeline_components/preview/model_evaluation/__init__.py

@@ -1,4 +1,4 @@
-# Copyright 2021 The Kubeflow Authors. All Rights Reserved.
+# Copyright 2024 The Kubeflow Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -17,10 +17,12 @@ from google_cloud_pipeline_components.preview.model_evaluation.data_bias_compone
 from google_cloud_pipeline_components.preview.model_evaluation.feature_attribution_component import feature_attribution as ModelEvaluationFeatureAttributionOp
 from google_cloud_pipeline_components.preview.model_evaluation.feature_attribution_graph_component import feature_attribution_graph_component as FeatureAttributionGraphComponentOp
 from google_cloud_pipeline_components.preview.model_evaluation.model_bias_component import detect_model_bias as DetectModelBiasOp
+from google_cloud_pipeline_components.preview.model_evaluation.model_evaluation_import_component import model_evaluation_import as ModelImportEvaluationOp
 from google_cloud_pipeline_components.v1.model_evaluation.evaluation_llm_classification_pipeline import evaluation_llm_classification_pipeline
 from google_cloud_pipeline_components.v1.model_evaluation.evaluation_llm_text_generation_pipeline import evaluation_llm_text_generation_pipeline
 from google_cloud_pipeline_components.v1.model_evaluation.model_based_llm_evaluation.autosxs.autosxs_pipeline import autosxs_pipeline
 
+
 __all__ = [
     'autosxs_pipeline',
     'evaluation_llm_classification_pipeline',
@@ -29,4 +31,5 @@ __all__ = [
     'FeatureAttributionGraphComponentOp',
     'DetectModelBiasOp',
     'DetectDataBiasOp',
+    'ModelImportEvaluationOp',
 ]

google_cloud_pipeline_components/{_implementation/model_evaluation/import_evaluation/component.py → preview/model_evaluation/model_evaluation_import_component.py}

@@ -1,4 +1,4 @@
-# Copyright 2023 The Kubeflow Authors. All Rights Reserved.
+# Copyright 2024 The Kubeflow Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -53,8 +53,9 @@ def model_evaluation_import(
 
   For more details, see
   https://cloud.google.com/vertex-ai/docs/reference/rest/v1/projects.locations.models.evaluations
-  One of the four metrics inputs must be provided, metrics & problem_type,
-  classification_metrics, regression_metrics, or forecasting_metrics.
+  One of the metrics inputs must be provided, metrics & problem_type,
+  classification_metrics, regression_metrics, or forecasting_metrics, text_generation_metrics,
+  question_answering_metrics, summarization_metrics, embedding_metrics.
 
   Args:
     model: Vertex model resource that will be the parent resource of the

google_cloud_pipeline_components/preview/starry_net/__init__.py

@@ -0,0 +1,19 @@
+# Copyright 2024 The Kubeflow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Starry Net Forecasting Pipeline."""
+
+from google_cloud_pipeline_components.preview.starry_net.component import starry_net  # pylint: disable=g-importing-member
+
+__all__ = ['starry_net']