Skip to content

Using Mlflow and Jaxtrain with --data-generation-experiment-id #73

Description

@hannahhafner

When I run jaxtrain without providing a training data folder and using the experiment ID parameter, it fails with this error:

ValueError: Failed to get training data from MLflow and no 
--training-data-folder provided: Could not determine training_data_folder from 
MLflow experiment 1. Please provide --training-data-folder explicitly.

Upon further inspection, I think the error comes from this line in jax_train:
264: mlflow_data_folder = first_run_info.get("data_output_folder")

In the first_run_info dictionary, I have no key for 'data_output_folder' even though I do have keys for 'run_id',`` 'run_name', 'num_files', 'total_size_mb' and 'files' . It looks like mlflow_lineage_info was created, as the logger message MLflow reports 2 runs with 90 total files was printed correctly, but the data_output_folder is not stored within first_run_info

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions