mlflow: [BUG] A problem with type checking for string objects (MLflow-deployed model in SageMaker)

Thank you for submitting an issue. Please refer to our issue policy for additional information about bug reports. For help with debugging your code, please refer to Stack Overflow.

Please fill in this bug report template to ensure a timely and thorough response.

Willingness to contribute

The MLflow Community encourages bug fix contributions. Would you or another member of your organization be willing to contribute a fix for this bug to the MLflow code base?

• Yes. I can contribute a fix for this bug independently.
• Yes. I would be willing to contribute a fix for this bug with guidance from the MLflow community.
• No. I cannot contribute a bug fix at this time.

System information

• Have I written custom code (as opposed to using a stock example script provided in MLflow): Yes
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): ‘Linux’, ‘4.14.252-131.483.amzn1.x86_64’
• MLflow installed from (source or binary): pip
• MLflow version (run mlflow --version): mlflow==1.22.0
• Python version: python=3.6.10
• npm version, if running the dev UI: -
• Exact command to reproduce: -

Describe the problem

I deployed a Huggingface Transformer model in SageMaker using MLflow’s sagemaker.deploy().

The model had been tested after training (using the same test example that was used in the code that led to the described bug).

When logging the model I used infer_signature(np.array(test_example), loaded_model.predict(test_example)) to infer input and output signatures.

Model is deployed successfully. When trying to query the model I get ModelError (full traceback below).

To query the model, I am using precisely the same test_example that I used for infer_signature():

test_example = [['This is the subject', 'This is the body']]

The only difference is that when querying the deployed model, I am not wrapping the test example in np.array() as that is not json-serializeable.

To query the model I tried two different approaches:

import json
import boto3
import pandas as pd

SAGEMAKER_REGION = 'us-west-2'
MODEL_NAME = '...'

client = boto3.client("sagemaker-runtime", region_name=SAGEMAKER_REGION)

test_example = [['This is the subject', 'This is the body']]

# Approach 1
client.invoke_endpoint(
                EndpointName=MODEL_NAME,
                Body=json.dumps(test_example),
                ContentType="application/json",
            )

# Approach 2
client.invoke_endpoint(
                EndpointName=MODEL_NAME,
                Body=pd.DataFrame(test_example).to_json(orient="split"),
                ContentType="application/json; format=pandas-split",
            )

but they result in the same error.

To check if the problem is not in the model itself or in other components, I built a simple workaround.

I encoded strings into numbers (using ord()) and then decoded them back to strings (using chr()) inside the model wrapper. This solved the issue.

Summarizing, the same code worked for integer data, but not for string data.

Code to reproduce issue

---

Other info / logs

Include any logs or source code that would be helpful to diagnose the problem. If including tracebacks, please include the full traceback. Large logs and files should be attached.

---------------------------------------------------------------------------
ModelError                                Traceback (most recent call last)
<ipython-input-89-d09862a5f494> in <module>
      2                 EndpointName=MODEL_NAME,
      3                 Body=test_example,
----> 4                 ContentType="application/json; format=pandas-split",
      5             )

~/anaconda3/envs/amazonei_tensorflow_p36/lib/python3.6/site-packages/botocore/client.py in _api_call(self, *args, **kwargs)
    393                     "%s() only accepts keyword arguments." % py_operation_name)
    394             # The "self" in this scope is referring to the BaseClient.
--> 395             return self._make_api_call(operation_name, kwargs)
    396 
    397         _api_call.__name__ = str(py_operation_name)

~/anaconda3/envs/amazonei_tensorflow_p36/lib/python3.6/site-packages/botocore/client.py in _make_api_call(self, operation_name, api_params)
    723             error_code = parsed_response.get("Error", {}).get("Code")
    724             error_class = self.exceptions.from_code(error_code)
--> 725             raise error_class(parsed_response, operation_name)
    726         else:
    727             return parsed_response

ModelError: An error occurred (ModelError) when calling the InvokeEndpoint operation: Received client error (400) from primary with message "{"error_code": "BAD_REQUEST", "message": "dtype of input object does not match expected dtype <U0"}". See https://us-west-2.console.aws.amazon.com/cloudwatch/home?region=us-west-2#logEventViewer:group=/aws/sagemaker/Endpoints/bec-sagemaker-model-test-app in account 543052680787 for more information.

Environment info:

{'channels': ['defaults', 'conda-forge', 'pytorch'],
 'dependencies': ['python=3.6.10',
  'pip==21.3.1',
  'pytorch=1.10.2',
  'cudatoolkit=10.2',
  {'pip': ['mlflow==1.22.0',
    'transformers==4.17.0',
    'datasets==1.18.4',
    'cloudpickle==1.3.0']}],
 'name': 'bert_bec_test_env'}

What component(s), interfaces, languages, and integrations does this bug affect?

Components

• area/artifacts: Artifact stores and artifact logging
• area/build: Build and test infrastructure for MLflow
• area/docs: MLflow documentation pages
• area/examples: Example code
• area/model-registry: Model Registry service, APIs, and the fluent client calls for Model Registry
• area/models: MLmodel format, model serialization/deserialization, flavors
• area/projects: MLproject format, project running backends
• area/scoring: MLflow Model server, model deployment tools, Spark UDFs
• area/server-infra: MLflow Tracking server backend
• area/tracking: Tracking Service, tracking client APIs, autologging

Interface

• area/uiux: Front-end, user experience, plotting, JavaScript, JavaScript dev server
• area/docker: Docker use across MLflow’s components, such as MLflow Projects and MLflow Models
• area/sqlalchemy: Use of SQLAlchemy in the Tracking Service or Model Registry
• area/windows: Windows support

Language

• language/r: R APIs and clients
• language/java: Java APIs and clients
• language/new: Proposals for new client languages

Integrations

• integrations/azure: Azure and Azure ML integrations
• integrations/sagemaker: SageMaker integrations
• integrations/databricks: Databricks integrations