IssueTracker

Problem you have encountered:

I'm trying to create a Vertex AI Batch Prediction Job using the Python SDK. I'm using a custom container with a Flask/gunicorn server that expects both instances and parameters fields in the prediction method logic based on the documentation in this link. I'm using the google.cloud.aiplatform Python SDK to create the Batch Prediction Job with a GCS JSONL file as input, and this populates the instances field in the prediction route of my flask app as expected. I was wondering how to populate the parameters field. I tried to use the model_parameters parameter of the batch_predict method of the google.cloud.aiplatform.models.Model class, but this doesn't seem to work. Is this functionality supported or not for Batch Predictions?

What you expected to happen:

Passing a dictionary to the model_parameters parameter of the google.cloud.aiplatform.models.Model. batch_predict method should populate the request_json.get("parameters") in my custom container's Flask prediction route, but this returns None instead. request_json.get("instances") returns my input file as expected.

Steps to reproduce:

Code Excerpt:

######## Calling Code #########
from google.cloud import aiplatform

model = aiplatform.Model("model_id", project="project_id", location="gcp_region")

batch_prediction_job = model.batch_predict(
    job_display_name="some_name",
    gcs_source="gs://path_to_jsonl_input_file",
    gcs_destination_prefix="gs://path_to_destination_bucket",
    model_parameters={},  # a dictionary
    machine_type="n1-standard-2",
    accelerator_type="NVIDIA_TESLA_T4",
    accelerator_count=1
)

######## Container Code ###########
request_json = request.get_json()

inputs = request_json.get("instances")  # Gives me the input from my input source

request_parameters = request_json.get("parameters")  # Returns null/None. Expect this to return the `model_parameters` parameter from the `batch_predict` method.

Other information (workarounds you have tried, documentation consulted, etc):