bug: Cant deploy a Langchain ragchain using Nemo guardrails as serving endpoint on Databricks CPU-environment. But i can deploy the exact same config on Databricks GPU-environment.

[Sodrberg]

Did you check docs and existing issues?

• I have read all the NeMo-Guardrails docs
• I have updated the package to the latest version before submitting this issue
• (optional) I have used the develop branch
• I have searched the existing issues of NeMo-Guardrails

Python version (python --version)

Python3.11

Operating system/version

Databricks serving endpoint

NeMo-Guardrails version (if you must use a specific version and not the latest

No response

Describe the bug

I have a ML-FLow Langchain rag-chain model that is using nemo guardrails. Im using input and output rails, as well as "blocked terms" functions as actions for input and output.
My problem is that i cant deploy the chain as a serving endpoint on Databricks CPU-environment. But i CAN deploy the exact same config on Databricks GPU-environment. If i remove the guardrails, i can deploy it without issues on the CPU-environment.

I have tried deploying just the simplest Hello World example from the Nemo docs as a CPU-only serving endpoint with success, but when querying the endpoint, i get "bad request". When deploying the exact config with GPU, the endpoint works as expected.

Why does the nemo package work on databricks GPU environment but not with CPU-only environment?

Steps To Reproduce

1. Create an MlFlow model out of the nemo "Hello World" example.
2. Deploy as serving endpoint on databricks using "workload type": "CPU"

Expected Behavior

The endpoint is successfully created and provides answers as expected.

Actual Behavior

The endpoint creation times out and fails.