title: the two pools of the target and the draft
source: manuals/sglang-101/capture/gpu/out-gb10/server-eagle3.log
lines: 27-29
lang: text
note: The two lines report the same token count with very different sizes.
---
[2026-10-17 15:20:23] KV Cache is allocated. dtype: torch.bfloat16, #tokens: 216775, K size: 9.92 GB, V size: 9.92 GB
[2026-10-17 15:20:23] Memory pool end. avail mem=34.44 GB
[2026-10-17 15:20:23] KV Cache is allocated. dtype: torch.bfloat16, #tokens: 216775, K size: 0.21 GB, V size: 0.21 GB
