This message was deleted.

Hi Ariel, that is more of a limitation on MLflow side and will need to wait for MLflows gpu support. However, you can find some examples in that doc showing how to workaround this issue by using MLflow with BentoML native runners that supports GPU

Hi @Ariel Zadok, since we last looked, the lack of GPU support is in the MLflow side. You can load back the original flavor, e.g. sklearn, with MLflow and save with BentoML as instructed in the documentation.

loaded_model = mlflow.sklearn.load_model(model_uri)
bentoml.sklearn.save_model("iris_clf", loaded_model)

You will be able to serve with GPU using BentoML’s framework API.

But to the best of my knowledge, mlflow does support GPU, I am currently using mlflow on databricks with GPU. What am I missing here?

@Ariel Zadok I think it only supports training with GPU

Thanks, I will check it out