title: the one warning that sets the request limit
source: manuals/sglang-101/capture/gpu/out-gb10/server-eagle3.log
lines: 4-4
lang: text
note: This line prints once at startup and scrolls away before the first request.
---
[2026-10-17 15:14:58] Max running requests is reset to 48 for speculative decoding. You can override this by explicitly setting --max-running-requests.
