You mean if you used a modern LLM of GPT-2-level quality. Vanilla transformers like GPT 2 are ridiculously inefficient in comparison.