{% extends "base.html" %} {% block main %}
Benchmarks

Runs

Every benchmark you ran, newest first.

New run
{% if active %}
running Run {{ active.run_id or "" }} is in progress{% if active.source %} (started from {{ active.source }}){% endif %}. {% if active.run_id %}Follow it{% endif %}
{% endif %} {% if runs %}
{% for r in runs %} {% set badge = {"done": "ok", "running": "running", "failed": "bad", "cancelled": "warn", "interrupted": "warn"}.get(r.status, "") %} {% endfor %}
RunStartedModelLabelStatus
{% set finished = runs | selectattr("status", "equalto", "done") | list %} {% if finished | length >= 2 %}
Compare before, and after
{% endif %} {% else %}

No benchmarks yet

Point kneebench at an OpenAI-compatible vLLM endpoint. It works out what the hardware allows, measures what you get, and explains the gap.

Start your first benchmark Check your setup

{% endif %} {% endblock %}