Hello,
Would it be possible to publish a model asset together with its training experiment (MLflow) when the model has been trained outside of Foundry?
I.e: Model trained externally → Upload model weight and training logs → Publish model asset of externally trained weights + experiment.
Best regards,
Hello,
A model asset can be published from weights file once it is wrapped in an adapter class as detailed here. This method supports creating an experiment and adding any hyperparameters or metrics. However, Foundry cannot automatically translate a MLflow experiment to a Foundry experiment and the metrics/hyperparameters that one cares about would need to be logged explicitly in the code.
Thanks
Julien
I have no issue logging experiments during training within a Jupyter Notebook (you can even get the MLFlow UI available by using one of the port originally used by Streamlit / TensorBoard 
)
But indeed my question was around “importing” an existing ML Flow logging from external and forwarding it to Foundry experiments during Model asset publication.