Documenting MLflow

The connector is used to import metadata from MLflow into Dataedo, including:

• machine learning models (RegisteredModel)
• model versions (ModelVersion)

Dataedo connects to an MLflow instance, retrieves the complete metadata structure, and imports it into Dataedo as documentation objects.

Additionally, in the Source Description field, the connector automatically generates:

• Model Card – for a model
• Model Version Card – for a specific version of a model

Both cards are rendered as HTML using dedicated templates embedded in the connector.

---

Structure of the Model Card Template (RegisteredModel)

The Model Card for a registered model includes the following sections:

1. Header

   • title (model name)
   • created by
   • created at
   • last updated
   • deployment ID
   • deployment status

2. Description

   • model description
   • or No description

3. Latest Versions

   • table containing:
     • Version
     • Status
     • Description

4. Tags

   • table of model tags:
     • Key
     • Value

5. Aliases

   • table of model aliases:
     • Alias
     • Version

Structure of the Model Version Card Template (ModelVersion)

The Model Version Card includes detailed metadata about a specific version of a model:

1. Header

   • model name
   • version
   • status
   • stage
   • created at
   • last updated
   • source
   • run ID
   • run link

2. Description

   • version description
   • or No description

3. Signature

   • Inputs (Name / Type / Shape / Required)
   • Outputs (Name / Type / Shape / Required)
   • If not available → No signature defined

4. Flavors

   • list of model flavors
   • e.g., python_function, sklearn
   • with flavor-specific metadata

5. Saved Input Example

   • table containing:
     • Artifact path
     • Pandas orient
     • Serving input path
     • Type

6. Technical Metadata

   • MLflow version
   • Model ID
   • Model UUID
   • Size (bytes)
   • UTC Created timestamp

7. Parameters

   • table of run parameters (Key / Value)

8. Metrics

   • table of metrics (Key / Value / Step)

---

Connecting to MLflow

Importing Metadata in Dataedo Portal

Entry point

caution

This action can be performed only by users with the Connection Manager role

Navigate to:

Connections → Add new connection → MLflow

This will open the import wizard.

Step 1. Connection details

info

A Connection in Dataedo represents a saved configuration for accessing a data source.
It can be reused for future imports and scheduling.

Provide the required connection details, including the Connection name.
You can also add an optional description to provide additional context.

Fill in the Host field by entering the address together with the port.

Step 2. Credentials

Choose your credentials from the list of the already saved ones, or add new ones using the New credentials button.

Step 3. Objects to import

For each selected database, you can choose which objects to import:

• Select object types.

• Use Advanced filters to include or exclude objects with name patterns

Step 4. Schedule

caution

You must configure at least one import task in the schedule section.
If you skip this, an empty database will be created and no metadata will be imported.

Configure scheduling options for each source individually:

• Define tasks you want to schedule (Metadata Import, Data Quality run, Refresh Profiling)
• Run daily, on selected weekdays, or on specific days of the month
• Choose an exact time of execution
• Task state:
  • Active – the task will run as scheduled
  • Draft – the task is saved but not executed until switched to Active
• Run immediately – when checked, the task will also be executed right after clicking Create connection.

useful tip

Only one source in a metadata import can have Run immediately selected.

---

Limitations

• Dataedo does not import Execution or Prompt objects. The metadata of Run objects is part of the Model Version Card documentation, but they are not separate objects.

Specification

Imported metadata

	Imported	Editable
Model	✅	✅
Model Card (Source Description)	✅
Model Version	✅	✅
Model Version Card (Source Description)	✅
Parameters	✅	✅
Inputs	✅	✅
Outputs	✅	✅