idle
No run selected
Paste an OpenAI-compatible endpoint, pick a model and run the benchmark. Results stream in live; finished runs are saved to History automatically.
Progress
0%Headline
Throughput vs input
no points yet
no points yet
no points yet
no points yet
Context ladder
| Context | Prompt tok | TTFT ms | Prefill tok/s | Decode tok/s | Total s | Draft accept | Status |
|---|
No context points yet.
Concurrency
| n | Agg tok/s | Per-req tok/s | TTFT p50 ms | TTFT p95 ms | Wall s | Completion tok | Status |
|---|
No concurrency points yet.
GPU telemetry
No GPU agent attached — telemetry unavailable.
GPU agent
not connected
—
Run this on the GPU host to stream VRAM / power / temp / utilisation.
Export
Log
History unavailable
| Label | Model | Endpoint | TTFT p50 | Decode | Prefill @4K | Agg @8 | GPU | Created | Status | Actions |
|---|
No runs yet
Run a benchmark on the Run tab — completed runs are saved here automatically.
Nothing to compare yet
Select two or more runs on the History tab and press Compare selected. The first selected run is the baseline; every other run is shown as a signed, coloured delta against it.