{% if error %}
{{ error }}
{% else %} {% if busy %}
The server is already handling {{ "%.0f"|format(busy.running) }} running and {{ "%.0f"|format(busy.waiting) }} waiting requests. Other traffic blurs every number, and this test adds load to whoever is using it. Benchmark a replica nobody else is using if you can.
{% endif %}

{{ plan.points | length }} points, about {{ "%.0f"|format(plan.eta_s / 60) }} minutes including about 8 s of AIPerf startup per point; the sweep usually stops early

{% for p in plan.points %} {% endfor %}
#PhasePromptOutputIn flightRequestsEstimate
{{ loop.index }}{{ p.phase }}{{ p.spec.isl }}{{ p.spec.osl }} {% if p.at_operating_point %}best load found{% elif p.rate_factor %}{{ "%.0f"|format(p.rate_factor * 100) }}% of capacity{% else %}{{ p.spec.concurrency }}{% endif %} {% if p.at_operating_point or p.rate_factor %}set at run time{% else %}{{ p.spec.request_count }}{% endif %}{{ "%.0f"|format(p.eta_s) }} s
{% if plan.dropped_for_budget %}

Dropped to fit the budget: {{ plan.dropped_for_budget | join(", ") }} users.

{% endif %} {% endif %}