DeepSeek-V4.1-Flash
connecting...
Decode
--
tok/s
Prefill
--
tok/s
Running
--
KV pool
--
%
Latency
--
ms/tok
Total generated
--
Throughput
decode
prefill
Concurrency & KV pool
running
queued
KV %
Per-concurrency statistics
reset
Running
Observed
Decode mean
Per stream
Peak decode
Prefill mean
ITL
no traffic observed yet
Speculative decoding
ACCEPT
--
TOK/DRAFT
--
TOK/STEP
--
acceptance by draft position
GPUs
Since boot