Service Orchestration not working with multiple HuggingFace models

[skel84]

Describe the bug

I created a Seldon Deployment with the following configuration

---
apiVersion: machinelearning.seldon.io/v1alpha2
kind: SeldonDeployment
metadata:
  name: test-deployment
spec:
  protocol: v2
  predictors:
  - graph:
      name: transformer
      modelUri: s3://xxxxxxxxxx
      implementation: HUGGINGFACE_SERVER
      envSecretRefName: seldon-init-container-secret
      logger:
        mode: all      
    name: default
    replicas: 1

The S3 bucket contains 2 models and they are correctly loaded by the main container. Trying to reach the model, I get this error:

curl -s http://localhost:60253/seldon/default/francesco/v2/models/distilgpt2
{"status":{"code":500,"info":"Failed to find model distilgpt2","status":"FAILURE"}}%

To reproduce

1. Deploy like above, from a location that includes multiple model definitions

Expected behavior

If I include the annotation seldon.io/no-engine: "true" to the deployment, it works

curl -s http://localhost:60253/seldon/default/francesco/v2/models/distilgpt2
{"name":"distilgpt2","versions":[],"platform":"","inputs":[],"outputs":[],"parameters":{"content_type":null,"headers":null}}%