MLR and vLLM conflict, code not runnable

[azaccor]

Hello, I know this is a bit old: llm-models/mistral/mistral-7b/01_b_load_inference_vllm.py, but I'm looking for example code snippets with VLLM working on DBR/MLR. Your markdown refers to MLR 14.0 GPU, which is no longer available, but even using 14.3 LTS ML GPU with the g5.4xlarge gives an error when trying to import vllm.

I've tried my hand at this and I can confirm that this:

%pip install mlflow==2.20.3 vllm==0.4.2 transformers==4.49.0 huggingface_hub==0.27.1 openai==1.65.4 cloudpickle==2.2.1
dbutils.library.restartPython()

Allows you to import vllm as LLM using it in 15.4 LTS ML GPU environment.

Do you have any currently operable examples of using VLLM on Databricks that I can reference?