4 ms·

Does this make it appear that the LLM's responses converge on one answer when actually it's just caching?