^Sorry. I already fixed that one. Turned out that unlike before (1.0.07a), you need to load it using

onnx.load(model_path)

I think my question is that we can do this anymore? CLI

bentoml serve classifier.py:service --reload                               <----------- for autoreload

Script

bentoml.onnx.save_model(f"{BENTO_MODEL}", onnx.load(model_path))           <----------- SAVE and Load model at script start

runner = bentoml.onnx.get(f"{BENTO_MODEL}:{MODEL_VERSION}").with_options(  <----------- Get running
    providers=["CPUExecutionProvider"]
).to_runner()

service = bentoml.Service(f"{BENTO_MODEL}", runners=[runner])


@service.api(input=JSON(), output=JSON())
   ...