Troubleshooting¶
Resolve walkthrough, Swagger, and prediction API access problems.
The walkthrough does not finish¶
Cause: cloud-init has not finished, or the shell is not ubuntu.
Fix:
Swagger does not load¶
Cause: sushi-bentoml-example is not running, or port 3000 is not
forwarded.
Fix: as ubuntu, run sushi-bentoml-example, then forward port 3000
and open it locally. In Swagger, pick an endpoint, Try it out, then
Execute. See Open Swagger.
The prediction API returns an authorization error¶
Cause: the call did not use the per-instance token. The listing says
sushi-bentoml-api reads the access file for you.
Fix: use that helper instead of pasting a token into the command. Do not print the token. Forward port 8448 if the client is on your workstation.
You expected a call to a hosted model¶
Cause: this image does not call a model provider. The first prediction uses the bundled scikit-learn sample.
Fix: run the sample first. Serving a different model is described in
/opt/sushi/docs/README.md and in the
BentoML documentation, not as a provider key
on this image.
Last updated November 1, 2026 · Suggest an improvement