Error in trying to register to Saved Model with prediction type 'OTHER'

georgeannie

Hi

I am trying to follow the MLOps Best practices to deploy a model in production. I am trying register a model after adding the model in experiment tracking. The model has been added in experiment tracking successfully as a 'OTHER' model using the below syntax

mlflow.pyfunc.log_model(

  artifact_path=f"model_age}",

  python_model=CustomWrapper(),

  artifacts={"model":  'model.pkl'}

)

mlflow_extension = project.get_mlflow_extension()

mlflow_extension.set_run_inference_info(run_id=run.info.run_id,

     prediction_type='OTHER',

     code_env_name='Custom')

Following MLOps, the intent is to compare the model logged in experiment tracking with the previous metric in saved model.

For testing I just want to register the model from experiment tracking to saved model. Here is the code I used:

SAVED_MODEL_ID = None

for sm in project.list_saved_models():

    if run_name != sm["name"]:

        continue

    else:

        SAVED_MODEL_ID = sm["id"]

        print(

            "Found SavedModel {} with id {}".format(

                run_name, SAVED_MODEL_ID))

        break

if SAVED_MODEL_ID:

    sm = project.get_saved_model(SAVED_MODEL_ID)

else:

    sm = project.create_mlflow_pyfunc_model(

        name=run_name,

        prediction_type="OTHER")


# Deploy MLflow model as a saved model version

if len(sm.list_versions()) == 0:

    version_id = "V1"

else:

    max_version_num = max([int(v['id'][1:]) for v in sm.list_versions()])

    version_id = f"V{max_version_num+1}"


# Create version in SavedModel

mlflow_extension.deploy_run_model(

    run_id=run.info.run_id,

    sm_id=sm.id,

    version_id=version_id

    )

The above code returns error -

DataikuException: com.dataiku.dip.io.SocketBlockLinkKernelException: Could not run command READ_META: : <class 'EOFError'> :

Would appreciate any leads to fix this error.

Operating system used: Windows

Operating system used: Windows

Operating system used: Windows