This message was deleted.

I've just tried the direct loading idea I mentioned: It doesn't work as it is not a pytorch.nn model. Bentoml may not have any premade runners that support it.