MlflowException: The specified model does not contain any of the supported flavors for deployment.

[jaykishan-b]

While deploying the model generated with the regression example, getting below error: (In this case, the model has been created, but getting error while deploying the model on sagemaker)

MlflowException: The specified model does not contain any of the supported flavors for deployment. The model contains the following flavors: dict_keys(['sklearn']). Supported flavors: ['python_function', 'mleap']

Script for model deployment: https://github.com/aws-samples/amazon-sagemaker-mlflow-fargate/blob/main/lab/3_deploy_model.ipynb

Full traceback:

---------------------------------------------------------------------------
MlflowException                           Traceback (most recent call last)
Cell In[11], line 11
      1 config={
      2     'execution_role_arn': role,
      3     'image_url': image_uri,
   (...)
      6     'region_name': region
      7 }
      9 client = get_deploy_client("sagemaker")
---> 11 client.create_deployment(
     12     name=endpoint_name,
     13     model_uri=model_uri,
     14     # flavor='sklearn',
     15     config=config
     16 )

File ~/.local/share/virtualenvs/recipes-examples-49tQ2II3/lib/python3.10/site-packages/mlflow/sagemaker/__init__.py:2206, in SageMakerDeploymentClient.create_deployment(self, name, model_uri, flavor, config, endpoint)
   2203 if config:
   2204     self._apply_custom_config(final_config, config)
-> 2206 app_name, flavor = _deploy(
   2207     app_name=name,
   2208     model_uri=model_uri,
   2209     flavor=flavor,
   2210     execution_role_arn=final_config["execution_role_arn"],
   2211     assume_role_arn=final_config["assume_role_arn"],
   2212     bucket=final_config["bucket"],
   2213     image_url=final_config["image_url"],
   2214     region_name=final_config["region_name"],
   2215     mode=mlflow.sagemaker.DEPLOYMENT_MODE_CREATE,
   2216     archive=final_config["archive"],
   2217     instance_type=final_config["instance_type"],
   2218     instance_count=final_config["instance_count"],
   2219     vpc_config=final_config["vpc_config"],
   2220     data_capture_config=final_config["data_capture_config"],
   2221     synchronous=final_config["synchronous"],
   2222     timeout_seconds=final_config["timeout_seconds"],
   2223     variant_name=final_config["variant_name"],
   2224     async_inference_config=final_config["async_inference_config"],
   2225     env=final_config["env"],
   2226     tags=final_config["tags"],
   2227 )
   2229 return {"name": app_name, "flavor": flavor}

File ~/.local/share/virtualenvs/recipes-examples-49tQ2II3/lib/python3.10/site-packages/mlflow/sagemaker/__init__.py:365, in _deploy(app_name, model_uri, execution_role_arn, assume_role_arn, bucket, image_url, region_name, mode, archive, instance_type, instance_count, vpc_config, flavor, synchronous, timeout_seconds, data_capture_config, variant_name, async_inference_config, env, tags)
    362 model_config = Model.load(model_config_path)
    364 if flavor is None:
--> 365     flavor = _get_preferred_deployment_flavor(model_config)
    366 else:
    367     _validate_deployment_flavor(model_config, flavor)

File ~/.local/share/virtualenvs/recipes-examples-49tQ2II3/lib/python3.10/site-packages/mlflow/sagemaker/__init__.py:68, in _get_preferred_deployment_flavor(model_config)
     66     return pyfunc.FLAVOR_NAME
     67 else:
---> 68     raise MlflowException(
...
     76         ),
     77         error_code=RESOURCE_DOES_NOT_EXIST,
     78     )

	MlflowException: The specified model does not contain any of the supported flavors for deployment. The model contains the following flavors: dict_keys(['sklearn']). Supported flavors: ['python_function', 'mleap']

Below is the MLmodel file:

artifact_path: transform/transformer
flavors:
  sklearn:
    code: code
    pickled_model: model.pkl
    serialization_format: cloudpickle
    sklearn_version: 1.3.0
mlflow_version: 2.6.0
model_uuid: 4d185e028de7487db974b36311d37d87
run_id: 57eb75a5c4a74f12b33d6d2146bc3bce
utc_time_created: '2023-08-24 09:15:53.090506'

Tried modifying estimator function in train.py step as below to get python_function flavor in the MLmodel file:

Reference: https://github.com/mlflow/mlflow/blob/master/docs/source/models.rst#scikit-learn-sklearn

def estimator_fn(estimator_params: Dict[str, Any] = None):
    """
    Returns an *unfitted* estimator that defines ``fit()`` and ``predict()`` methods.
    The estimator's input and output signatures should be compatible with scikit-learn
    estimators.
    """
    if estimator_params is None:
        estimator_params = {}

    sgd_regressor = SGDRegressor(random_state=42, **estimator_params)
    
    model_info = mlflow.sklearn.log_model(
        sk_model=sgd_regressor, artifact_path="sklearn_pyfunc_model_artifact"
    )

    sklearn_pyfunc = mlflow.pyfunc.load_model(model_uri=model_info.model_uri)
    return sklearn_pyfunc`

With above modification, getting below error (In this case, getting error while running train step of the recipe):

MlflowException                           Traceback (most recent call last)
Cell In[17], line 1
----> 1 r.run("train")

File ~/.local/share/virtualenvs/recipes-examples-49tQ2II3/lib/python3.10/site-packages/mlflow/recipes/regression/v1/recipe.py:276, in RegressionRecipe.run(self, step)
    204 def run(self, step: str = None) -> None:
    205     """
    206     Runs the full recipe or a particular recipe step, producing outputs and displaying a
    207     summary of results upon completion. Step outputs are cached from previous executions, and
   (...)
    274         regression_recipe.run()
    275     """
--> 276     return super().run(step=step)

File ~/.local/share/virtualenvs/recipes-examples-49tQ2II3/lib/python3.10/site-packages/mlflow/recipes/recipe.py:98, in BaseRecipe.run(self, step)
     96 if last_executed_step_state.status != StepStatus.SUCCEEDED:
     97     if step is not None:
---> 98         raise MlflowException(
     99             f"Failed to run step '{step}' of recipe '{self.name}'."
    100             f" An error was encountered while running step '{last_executed_step.name}':"
    101             f" {last_executed_step_state.stack_trace}",
    102             error_code=BAD_REQUEST,
    103         )
    104     else:
    105         raise MlflowException(
    106             f"Failed to run recipe '{self.name}'."
    107             f" An error was encountered while running step '{last_executed_step.name}':"
    108             f" {last_executed_step_state.stack_trace}",
    109             error_code=BAD_REQUEST,
    110         )

MlflowException: Failed to run step 'train' of recipe 'regression'. An error was encountered while running step 'train': Traceback (most recent call last):
  File "/home/jaykishan/.local/share/virtualenvs/recipes-examples-49tQ2II3/lib/python3.10/site-packages/mlflow/recipes/step.py", line 129, in run
    self.step_card = self._run(output_directory=output_directory)
  File "/home/jaykishan/.local/share/virtualenvs/recipes-examples-49tQ2II3/lib/python3.10/site-packages/mlflow/recipes/steps/train.py", line 367, in _run
    estimator = self._resolve_estimator(
  File "/home/jaykishan/.local/share/virtualenvs/recipes-examples-49tQ2II3/lib/python3.10/site-packages/mlflow/recipes/steps/train.py", line 695, in _resolve_estimator
    return self._get_user_defined_estimator(
  File "/home/jaykishan/.local/share/virtualenvs/recipes-examples-49tQ2II3/lib/python3.10/site-packages/mlflow/recipes/steps/train.py", line 665, in _get_user_defined_estimator
    default_params = estimator.get_params()
AttributeError: 'PyFuncModel' object has no attribute 'get_params'

Any idea about this?

[benjaminbascary]

Any advance on this? I got stuck in the same part