QGRE chat — live moat demo

QGRE▙▟

—tok/s

modelloading…

tok/s—

reqs—

queue—

offline

[settings]sampling

knobs forwarded to qgre-server via prepareCall. applied per-request; persists to sessionStorage across refreshes.

temperature0.70

[0.00]0 = greedy · 0.7 default · >1.5 chaotic[2.00]

top_p0.80

[0.00]nucleus size · qwen3 default 0.8[1.00]

top_k20

[1]vocabulary clamp · qwen3 default 20[100]

qwen3 recipe · session-scoped

[history]—

loading…

throughput

—tok/s

warmup — tok/s

kv prefix reuse

—

tokens not re-prefilled

total generated

—

tokens this session

quant—

max seq—

recipe—

prompt >

⌘↵ send