There is a package conflic when Responsible AI is trying to run. How do I resolve this?

Question

There is a package conflic when Responsible AI is trying to run. How do I resolve this?

Tariel Chichua 0

I am using Azure AutoML to train a predictive model that can guess segments. One of the child jobs for Responsible AI fails with the error below. Because of this my trained model does not have an explanation. I need to resolve this, so that I can understand which variables are good predictors in my ML model.

Execution failed. User process 'python' exited with status code 1. Please check log file 'user_logs/std_log.txt' for error details. Error: 
The conflict is caused by:
    The user requested spacy==3.7.4
    en-core-web-sm 3.4.1 depends on spacy<3.5.0 and >=3.4.0

To fix this you could try to:
1. loosen the range of package versions you've specified
2. remove package versions to allow pip attempt to solve the dependency conflict

Pip subprocess error:
ERROR: Cannot install -r /mnt/azureml/cr/j/599bc9df29c54a5ea77db667ea120275/exe/wd/condaenv.gxdbprd_.requirements.txt (line 75) and spacy==3.7.4 because these package versions have conflicting dependencies.
ERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/topics/dependency-resolution/#dealing-with-dependency-conflicts


failed

CondaEnvException: Pip failed

Manas Mohanty 3,700 Reputation points Microsoft External Staff Moderator

2025-05-01T10:28:25.5866667+00:00

Hi Tariel Chichua

Good day.

AutoML experiment creates an environment automatically. We might be able to fix the package conflict by unpinning the package version from requirement.txt for below.

Please locate the "gxdbprd_.requirements.txt" from environment of respective job

and replace "spacy" with "spacy==3.7.4" and resubmit the job.

I am trying to replicate the scenario with below sample notebook

Samples/SDK v2/sdk/python/jobs/automl-standalone-jobs/automl-image-instance-segmentation-task-fridge-items

Also, would request you to share any other example document, if possible, to replicate and resolve the issue.

Thank you.
Manas Mohanty 3,700 Reputation points Microsoft External Staff Moderator

2025-05-02T08:56:46.3033333+00:00

Hi Tariel Chichua

We are checking to see if the unpinning the package version in requirement.txt of respective environment for failed job fixed the issue.

Thank you.
Tariel Chichua 0 Reputation points

2025-05-05T17:13:20.8633333+00:00

Thank you, I will attempt this asap
Tariel Chichua 0 Reputation points

2025-05-05T17:20:15.49+00:00

I went to the environment that the job was calculated in and this is what I see. It's not loading the information:
Tariel Chichua 0 Reputation points

2025-05-05T17:26:48.3966667+00:00

Also this is what I see in that environment on Jobs tab. The first Responsible AI task failed because of timeout. It was running on a slow CPU compute. Then I switched to a GPU compute and it failed after 9 minutes because of the above error.
JAYA SHANKAR G S 2,525 Reputation points Microsoft External Staff Moderator

2025-05-06T13:08:39.5133333+00:00

Hello @Tariel Chichua ,

Currently AutoML models are not supported for generating responsible AI dashboards.

Here is the limitation and supported models and scenarios, please refer this.

Thank you
Tariel Chichua 0 Reputation points

2025-05-06T15:42:48.98+00:00

Ok I understand, but you have removed model explanation from AutoML and moved it into Responsible AI, which is not supported in AutoML. How am I supposed to explain and understand the trained model now?

1 answer

Your answer

Manas Mohanty 3,700 Reputation points Microsoft External Staff Moderator

2025-05-01T10:28:25.5866667+00:00

Hi Tariel Chichua

Good day.

AutoML experiment creates an environment automatically. We might be able to fix the package conflict by unpinning the package version from requirement.txt for below.

Please locate the "gxdbprd_.requirements.txt" from environment of respective job

and replace "spacy" with "spacy==3.7.4" and resubmit the job.

I am trying to replicate the scenario with below sample notebook

Samples/SDK v2/sdk/python/jobs/automl-standalone-jobs/automl-image-instance-segmentation-task-fridge-items

Also, would request you to share any other example document, if possible, to replicate and resolve the issue.

Thank you.
Manas Mohanty 3,700 Reputation points Microsoft External Staff Moderator

2025-05-02T08:56:46.3033333+00:00

Hi Tariel Chichua

We are checking to see if the unpinning the package version in requirement.txt of respective environment for failed job fixed the issue.

Thank you.
Tariel Chichua 0 Reputation points

2025-05-05T17:13:20.8633333+00:00

Thank you, I will attempt this asap
Tariel Chichua 0 Reputation points

2025-05-05T17:20:15.49+00:00

I went to the environment that the job was calculated in and this is what I see. It's not loading the information:
Tariel Chichua 0 Reputation points

2025-05-05T17:26:48.3966667+00:00

Also this is what I see in that environment on Jobs tab. The first Responsible AI task failed because of timeout. It was running on a slow CPU compute. Then I switched to a GPU compute and it failed after 9 minutes because of the above error.
JAYA SHANKAR G S 2,525 Reputation points Microsoft External Staff Moderator

2025-05-06T13:08:39.5133333+00:00

Hello @Tariel Chichua ,

Currently AutoML models are not supported for generating responsible AI dashboards.

Here is the limitation and supported models and scenarios, please refer this.

Thank you
Tariel Chichua 0 Reputation points

2025-05-06T15:42:48.98+00:00

Ok I understand, but you have removed model explanation from AutoML and moved it into Responsible AI, which is not supported in AutoML. How am I supposed to explain and understand the trained model now?

Answer 1

Hello @Tariel Chichua

As per this documentation, i believe your automl job is on regression or classification (binary and multi-class) models trained on tabular structured data.

Next, you register the model from best run.

User's image

Got to the model, check in MLmodel file under artifacts tab whether it has sklearn flavor and also download the zip file and extract it to folder.

User's image

If you find sklearn flavor, go to notebooks tab , upload the folder downloaded earlier like below.

User's image

Execute below code to register new model with sklearn flavor.

import mlflow
from azure.ai.ml.entities import Model
from azure.ai.ml.constants import AssetTypes


ml_client = MLClient(
    credential=DefaultAzureCredential(),
    subscription_id='subscription_id',
    resource_group_name='resource_group',
    workspace_name='workspace_name'
)

tmp_model = mlflow.sklearn.load("3hBestmodel")
mlflow.sklearn.save_model(path="custom_mlflow_sklearn_model",sk_model= tmp_model)

mlflow_model = Model(
    path="custom_mlflow_sklearn_model",
    type=AssetTypes.MLFLOW_MODEL,
    name="custom_mlflow_sklearn_model",
    description="MLflow model created from local path",
)

ml_client.models.create_or_update(mlflow_model)

Above code will register new mlflow model with only sklearn flavor.

Now got to the registered model and you can create dashboard.

User's image

Basically, in MLmodel file below content should not be there especially the azureml.engine as automl.

metadata:
  azureml.base_image: mcr.microsoft.com/azureml/curated/ai-ml-automl:12
  azureml.engine: automl

If you don't find sklearn flavor, then you need to follow steps given here to convert the model to sklearn flavor.

So, above are the workaround for automl models, if you have any query, please let us know in comments or in private message.

Note: Make sure you are having all dependencies for model.

Thank you

JAYA SHANKAR G S 2,525 Reputation points Microsoft External Staff Moderator

2025-05-08T10:19:45.5566667+00:00

Hello @Tariel Chichua,

Have you checked above workaround? please check it and let us know if you have any query.

Thank you
JAYA SHANKAR G S 2,525 Reputation points Microsoft External Staff Moderator

2025-05-09T04:25:52.32+00:00

Hello @Tariel Chichua,

I believe the above given workaround helped you, please do let us know if have any query

Thank you

Share via

There is a package conflic when Responsible AI is trying to run. How do I resolve this?

1 answer

Your answer