This ZeroGPU demo keeps local vLLM disabled. Benchmarking runs against a remote or external OpenAI-compatible endpoint.
Exports appear after the first successful run.