topic Re: Problems registering models via /api/2.0/mlflow/model-versions/create in Get Started Discussions

Problems registering models via /api/2.0/mlflow/model-versions/create

unj1m — Thu, 02 Jan 2025 22:05:15 GMT

Initially, I tried registering while logging a model from mlflow and got:

"Got an invalid source 'dbfs:/Volumes/compute_integration_tests/default/compute-external-table-tests-models/test_artifacts/1078d04b4d4b4537bdf7a4b5e94e9e7f/artifacts/model'. Only DBFS locations are currently supported."

WAT?

I got this when trying to log a model and supply a model name. I can also reproduce by calling the REST API directly.

After some trial and error, I found that using the source:

'dbfs:/dbfs/Volumes/compute_integration_tests/default/compute-external-table-tests-models/test_artifacts/1078d04b4d4b4537bdf7a4b5e94e9e7f/artifacts/model'

"worked".

But a new version never got created.

Also, the documentation doesn't mention the required run_id and source parameters.

Re: Problems registering models via /api/2.0/mlflow/model-versions/create

Alberto_Umana — Thu, 02 Jan 2025 22:15:12 GMT

Hi @unj1m,

The path should be in the format dbfs:/Volumes/... without the extra /dbfs prefix. The correct path should be: dbfs:/Volumes/compute_integration_tests/default/compute-external-table-tests-models/test_artifacts/1078d04b4d4b4537bdf7a4b5e94e9e7f/artifacts/model

Re: Problems registering models via /api/2.0/mlflow/model-versions/create

fmadeiro — Fri, 03 Jan 2025 00:29:16 GMT

The error happens because of how the source path is interpreted. Use the correct DBFS path format: dbfs:/Volumes/.... Avoid adding /dbfs redundantly, as it's already the DBFS root.

Key Fixes:

New Version Not Created: Ensure each log_model call uses a unique run_id and updated model metadata to trigger versioning.
Required Parameters: For REST API, include run_id (from the MLflow run) and source (DBFS path to the model artifact).

Example Payload:

{

"name": "my_model_name",

"source": "dbfs:/Volumes/my-model-path/model",

"run_id": "1234567890abcdef"

}

Re: Problems registering models via /api/2.0/mlflow/model-versions/create

unj1m — Fri, 03 Jan 2025 19:03:21 GMT

I know, right! 😉

Look at the error message I posted:

Adding the /dbfs was necessary to make it work.