feat(components): Implement new output format of inference component

PiperOrigin-RevId: 597621035

components/google-cloud/google_cloud_pipeline_components/_implementation/model_evaluation/model_inference/component.py

@@ -42,7 +42,6 @@ def model_inference_component_internal(
     request_params: Dict[str, Any] = {},
     max_request_per_minute: float = 3,
     max_tokens_per_minute: float = 10000,
-    target_field_name: str = '',
     query_field_name: str = '',
     display_name: str = 'third-party-inference',
     machine_type: str = 'e2-highmem-16',
@@ -68,11 +67,6 @@ def model_inference_component_internal(
       max_request_per_minute: Maximum number of requests can be sent in a
         minute.
       max_tokens_per_minute: float = 10000,
-      target_field_name: The full name path of the features target field in the
-        predictions file. Formatted to be able to find nested columns, delimited
-        by `.`. Alternatively referred to as the ground truth (or
-        ground_truth_column) field. If not set, defaulted to
-        `inputs.ground_truth`.
       query_field_name: The full name path of the features prompt field in the
         request file. Formatted to be able to find nested columns, delimited by
         `.`. Alternatively referred to as the ground truth (or
@@ -115,7 +109,7 @@ def model_inference_component_internal(
       custom_job_payload=utils.build_custom_job_payload(
           display_name=display_name,
           machine_type=machine_type,
-          image_uri=version.LLM_EVAL_IMAGE_TAG,
+          image_uri=version.LLM_EVAL_IMAGE_TAG,  # for local test and validation, use _IMAGE_URI.
           args=[
               f'--3p_model_inference={True}',
               f'--project={project}',
@@ -127,7 +121,6 @@ def model_inference_component_internal(
               f'--client_api_key_path={client_api_key_path}',
               f'--max_request_per_minute={max_request_per_minute}',
               f'--max_tokens_per_minute={max_tokens_per_minute}',
-              f'--target_field_name={target_field_name}',
               f'--query_field_name={query_field_name}',
               f'--gcs_output_path={gcs_output_path.path}',
               '--executor_input={{$.json_escape[1]}}',
@@ -150,7 +143,6 @@ def model_inference_component(
     inference_platform: str = 'openai_chat_completions',
     model_id: str = 'gpt-3.5-turbo',
     request_params: Dict[str, Any] = {},
-    target_field_name: str = '',
     query_field_name: str = 'prompt',
     max_request_per_minute: float = 3,
     max_tokens_per_minute: float = 10000,
@@ -174,11 +166,6 @@ def model_inference_component(
       inference_platform: Name of the inference platform.
       model_id: Name of the model to send requests against.
       request_params: Parameters to confirgure requests.
-      target_field_name: The full name path of the features target field in the
-        predictions file. Formatted to be able to find nested columns, delimited
-        by `.`. Alternatively referred to as the ground truth (or
-        ground_truth_column) field. If not set, defaulted to
-        `inputs.ground_truth`.
       query_field_name: The full name path of the features prompt field in the
         request file. Formatted to be able to find nested columns, delimited by
         `.`. Alternatively referred to as the ground truth (or
@@ -234,7 +221,6 @@ def model_inference_component(
       max_tokens_per_minute=max_tokens_per_minute,
       display_name=display_name,
       query_field_name=query_field_name,
-      target_field_name=target_field_name,
       machine_type=machine_type,
       service_account=service_account,
       network=network,
@@ -341,7 +327,6 @@ def model_inference_and_evaluation_component(
       max_request_per_minute=max_request_per_minute,
       max_tokens_per_minute=max_tokens_per_minute,
       query_field_name=query_field_name,
-      target_field_name=target_field_name,
       display_name=display_name,
       machine_type=machine_type,
       service_account=service_account,
@@ -354,6 +339,7 @@ def model_inference_and_evaluation_component(
       project=project,
       location=location,
       evaluation_task='text-generation',
+      target_field_name='.'.join(['instance', str(target_field_name)]),
       predictions_format='jsonl',
       joined_predictions_gcs_source=inference_task.outputs['gcs_output_path'],
       machine_type=machine_type,