BentoML

Looks like this has to do with how you are calling the runner's method?

also, is there any reasons why are you using our internal apis?

I'm invoking it with `logits_tensor = await runner.async_run(batch)`, being `batch` a `Dict[str, torch.Tensor]`

I'm using the internal APIs because I didn't want to pass just the name of the model as it's being done in the example in the docs. I'm already invoking `bentoml.pytorch.get()`  in the `service.py` because I need to read some `custom_objects` from it. So I thought it would be better to pass the instance I already have. Maybe it was not really necessary? Is there actually any downside of having two instances of a BentoModel in memory?

&gt; Is there actually any downside of having two instances of a BentoModel in memory
it takes up more memory? :slightly_smiling_face: otherwise there's not really a downside.

It's recommended to use our public API unless you really know what you're doing! have you tested with the public API instead?

anyway your original error has to do with
`def __call__(self, **batch):`

you should do this
`await runner.async_run(**batch)`