This is my output

I recommend setting save_io = True in your config (in components.llm) and investigating your raw model output. OpenLLaMA 3b is a comparatively old and small model, so it's quite possible that it returns wrong/non-parseable outputs.