This message was deleted.

@kurniarahmatt hmmm you shouldn't encounter this error. Let us dig in more and get back to you

oke thanks @Bo

Did you setup your aws credentials correctly? The errors fails at boto3 unable to call

ListObjects

right?

yes, I've been added aws cred on ~/.aws/credentials for s3 minio storage mlflow artifact

@Aaron Pham Could boto3 cause this error?

UnboundLocalError: local variable 'local_path' referenced before assignment

that errors resolved from the boto3 exception

alright, I'll check again on boto3 and mlflow correctly. But I have another problem here:

error  Traceback (most recent call last):                                                                                                                         
  File "/home/ai_platform/serving/src/create_bentos.py", line 51, in <module>                                                                                     
    bento_model = bentoml.mlflow.import_model(model_name, model_uri,                                                                                              
  File "/opt/anaconda3/envs/training/lib/python3.8/site-packages/bentoml/_internal/frameworks/mlflow.py", line 222, in import_model                               
    raise BentoMLException(f'artifact "{model_uri}" is not a MLflow model')                                                                                       
bentoml.exceptions.BentoMLException: artifact "<s3://ai-mlflow-artifacts/15/00f407db952e4a24ab060f04f1910572/artifacts>" is not a MLflow model

The error appears when the model has been registered in MLflow Registry via the Python API, but this error does not appear if the model registration is done through the MLflow UI. I'm really confused, what should I do. Thank you.

the path you gives is not a valid MLflow model, because that path doesn’t include a MLmodel file

Ok, I'll check again, and I have one more question: can I change path of bentos & models from ~/bentoml/bentos or ~/bentoml/models to other path or s3 storage? Thanks

@kurniarahmatt it’s not currently supported but should be fairly easy to add that support. For now, you may use the

export/impoart

API to export the models/bentos to s3 after it’s saved to local store