09-03-2026, 05:24 PM
That matches what I was seeing. I ended up in about the same place using vLLM. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.
throughput matters, but tail latency matters too
