Even with a very simple question such as “What is 5 + 5?”, the response takes noticeably long to appear. The model generates the correct answer, “5 + 5 = 10,” at around 160 tokens per second, but the total response time is approximately 1.24 seconds, with about 0.8 seconds before the first token is produced.
This seems like a significant amount of overhead for such a straightforward query. Is this expected behavior, or could there be an initialization or processing step causing the delay?