Chat with sushi
Dashboard
Connecting...ChartsTick the ones to show. Drag a chart by its strip, or use its arrows, to reorder.
-
Live statistics are off
This server is not collecting live statistics, so the speed tiles and charts have nothing to show. Turn them on to start.
Generation speed
-
Prompt speed
-
Memory
-
Prompt cache
-
Requests
-
Gen speed
-
Prompt speed
-
Requests
-
running / waiting
GPU
-
busy
Memory
-
used by sushi
First token
-
average wait
Cache reuse
-
of prompt tokens
Served
-
output tokens
Sessionsreused from cachenew this request
| ID | Model | Step | Context | Reused | Generated | Elapsed | Speed | Memory |
|---|
Prompt speed (tokens/s)one point per finished request
Generation speed (tokens/s)while generating
Requestsgeneratingreading promptwaitinganswered together
Response timefirst tokenlast token
GPU speed and loadbusyclock speed
Memory (GiB)in usereusable cacheothermachine memory
Prompt cache reusereused from cacheread now
-
Prompt cache storagememory and SSD
Draft acceptanceone point per request that drafted
Prompt and generated lengthpromptgenerated
Request outcomessince the server started
Why requests weren't batchedsteps a request ran alone while others ran
Prompt cache writesSSD, since the server started
A file written again counts again; this does not measure SSD wear.
Tokens verified per roundDSpark, rounds by tokens verified
Models
Sushi packs on Hugging Face
Settings
Storage
Prompt cache (SSD)
| Model | Size | Last used | Entries | ID | Share used | Written | Write rate |
|---|
Unused: no loaded model uses it (an older sushi version, a changed model, or a model that is not loaded).
Inference
Thinking effort
What each effort word a request names does on this family: whether it is offered, what the chat template receives and how many thinking tokens it may spend. Changes apply to the next request.
| Word | Offered | Sends | Cap | Default |
|---|
Server
Server port
Memory
Automatic update checks
About
Version-