Thank you! I think there is a lot to dive deep later with different hardware, inference engine, prompt/harness setup etc.