YataiService doesn't allow a BentoService without artifacts

[withsmilo]

Describe the bug

In order to avoid #1037 and #1065, I override the __init__ function of BentoService to download and initialize the required model files from an external service(MLflow). In this case, I don't need to pack any artifacts, but I found a problem in YataiService that doesn't allow this case. The current YataiService includes the assumption that BentoService must always have artifacts.

To Reproduce

from sklearn import svm
from sklearn import datasets
iris = datasets.load_iris()
X, y = iris.data, iris.target
clf = svm.SVC(gamma='scale')
clf.fit(X, y)

# test_bentoservice.py
import pandas as pd
from bentoml import env, artifacts, api, BentoService
from bentoml.adapters import DataframeInput

@env(infer_pip_packages=True)
class TestBentoService(BentoService):
    def __init__(self):
        # Download model files from MLflow and Initialize them
        pass
    @api(input=DataframeInput(), batch=True)
    def predict(self, df: pd.DataFrame):
        return None # We can predict something with model files here.

from test_bentoservice import TestBentoService
bento_service = TestBentoService()
saved_path = bento_service.save()

4. Run a YataiService and select the BentoService.

Expected behavior

YataiService should show even BentoService without artifacts.

Screenshots/Logs

Environment:

• OS: MacOS 11.2.2
• Python Version: 3.7.9
• BentoML Version: 0.11.0

Additional context

[yubozhao]

Run a YataiService and select the BentoService.

Is the selection happens in the web UI?

[withsmilo]

@yubozhao
Yes. When I select the BentoService without artifacts in YataiService UI, nothing is visible.

[yubozhao]

Got it. I will work on a fix. Thank you for finding it!