{% extends "base.html" %} {% block title %}Run · FreeWeight{% endblock %} {% block content %} {% if error %}
{{ run.id }}
{{ run.status }}
· {{ run.suite_key }} {{ run.suite_version }}
· {{ run.model_canonical_id }}
{% if run.label %} · {{ run.label }}{% endif %}
{% if run.suite_key.startswith('goal.') and run.status == 'completed' %}
· grade its human criteria
{% endif %}
{{ run.error_code }}
{{ run.error_text }}
Created {{ run.created_at | timestamp }} · started {{ run.started_at | timestamp }} ·
completed {{ run.completed_at | timestamp }}
Fingerprint {{ run.reproducibility_fingerprint }}
This run was measured under conditions worth knowing about.
{{ degradation.kind }} — {{ degradation.detail | tojson }}| Served context | {% if run.served_context is none %} — {% else %}{{ run.served_context }} tokens ({{ run.served_context_source }}){% endif %} |
|---|---|
| Attributed device | GPU {{ run.gpu_index if run.gpu_index is not none else "—" }} {% if run.multi_gpu_visible %} · more than one GPU was visible; memory, KV and energy figures are unsupported unless the provider reports placement {% endif %} |
| Telemetry sampling overhead | {% if run.telemetry_overhead_percent is none %} — {% else %}{{ "%.3f" | format(run.telemetry_overhead_percent) }} %{% endif %} |
| Prompt pack |
{% if run.prompt_pack_id %}
{{ run.prompt_pack_id }} {{ run.prompt_pack_version }}
{{ run.prompt_pack_hash[:19] }}…
provenance only — the fingerprint takes each benchmark's own prompt
subset hash
{% else %}This suite renders no prompt records.{% endif %}
|
The inputs the fingerprint was taken over, stored in full. A hash that cannot be explained is no use during a regression hunt.
{{ run.fingerprint_document | tojson(indent=2) }}
Waiting for events.
| Test | Status | Cases | Repetitions | Skip reason | Error |
|---|---|---|---|---|---|
| {{ test.test_name }} {{ test.test_key }} | {{ test.status }} | {{ test.completed_cases }} / {{ test.total_cases }} | {{ test.repetitions }} | {{ test.skip_reason or "—" }} | {{ test.error_code or "—" }} |
| No tests enumerated yet. | |||||
| Metric | Scope | GPU | Value | Unit | Aggregation | Spread | Samples | Excluded | Raw source |
|---|---|---|---|---|---|---|---|---|---|
| {{ metric.metric_key }} | {{ "run" if metric.run_test_id is none else "test" }} | {{ metric.gpu_index if metric.gpu_index is not none else "—" }} | {% if metric.unavailable_reason %} — {% else %} {{ "%.4f" | format(metric.numeric_value) }} {% endif %} | {{ metric.unit }} | {{ metric.aggregation }} | {% if metric.stddev is none %} — {% else %}± {{ "%.4f" | format(metric.stddev) }}{% endif %} | {{ metric.sample_count if metric.sample_count is not none else "—" }} | {{ metric.excluded_count if metric.excluded_count is not none else "—" }} | {# UI standards §5: at most two clicks from a headline metric to the raw record behind it. A test-scope metric links straight at its samples; a run-scope one combines every test, so it links at the tests table, where each row links at its own samples. Either way the second click lands on stored rows. #}{% if metric.run_test_id %} samples {% else %} every test {% endif %} |
| No aggregates yet — they are written once every sample is durable. | |||||||||
{{ telemetry_samples }} observation(s) recorded while this run executed. Each series names its device; nothing is summed or averaged across GPUs. A gap in a line is a reading this machine could not take, never a zero.
{% for chart in charts %}Axis 0 – {{ "%.2f" | format(chart.maximum) }} {{ chart.unit }} · min {{ "%.2f" | format(chart.minimum) }} · mean {{ "%.2f" | format(chart.mean) }} · max {{ "%.2f" | format(chart.maximum) }} {{ chart.unit }} · {{ chart.reported }} reported, {{ chart.missing }} not readable
No telemetry was recorded for this run.
Either this machine reports none, or telemetry.persist_during_runs is off. It is
real and empty, not missing.