LLM Monitor

poll 1000ms · read-only

--:--:--
all engines 2 engines · no data yet sums only engines answering now
running
n/a
waiting
n/a
prompt tok/s
n/a
decode tok/s
n/a
total tokens
n/a
gufo –
KV cache n/a
prompt tok/s
n/a
decode tok/s
n/a
total tokens
n/a
slots busy
n/a
prompt cache
n/a
spec accept
n/a
peak ctx
n/a
gpu losses
n/a
vLLM –
GPU cache n/a
prompt tok/s
n/a
decode tok/s
n/a
total tokens
n/a
per-seq tok/s
n/a
prefill avg
n/a
gufo no data yet
decode tok/s · all slots · last 60n/a
n/a n/a n/a
KV cache usage · last 60n/a
n/a n/a n/a

sessions (GET /slots)

waiting for /api/gufo/slots…

vLLM no data yet
decode tok/s · all slots · last 60n/a
n/a n/a n/a
GPU cache usage · last 60n/a
n/a n/a n/a

engine metrics (GET /metrics)

waiting · capacity

n/a

reason breakdown n/a

GPU cache usage

n/a

TTFT (mean)

n/a

p95 n/a

TPOT (mean)

n/a

p95 n/a

queue time (mean)

n/a

p95 n/a

prompt tokens

n/a

vllm:prompt_tokens_total

generation tokens

n/a

vllm:generation_tokens_total

prefix cache hit

n/a

vllm:prefix_cache_hits/queries

preemptions

n/a

running reqs evicted by KV pressure

series scraped

n/a

–

vLLM exposes no per-slot JSON — the metrics above are its equivalent of the session grid.