festbench-web/app/Console/Commands/SessionSummary.php

186 lines
7.2 KiB
PHP

<?php
namespace App\Console\Commands;
use App\Support\Suites;
use Illuminate\Console\Command;
use Illuminate\Support\Facades\DB;
use Symfony\Component\Console\Attribute\AsCommand;
#[AsCommand(name: 'festbench:summary', description: 'Rebuild the session-grouped home summary tables')]
class SessionSummary extends Command
{
public function handle(): int
{
$sessions = [];
$runs = DB::table('benchmark_runs')
->leftJoin('users', 'users.id', '=', 'benchmark_runs.user_id')
->orderByDesc('ran_at')
->orderByDesc('benchmark_runs.id')
->get([
'benchmark_runs.*',
'users.name as runner_name',
]);
foreach ($runs as $run) {
// Visibility participates in the grouping key so private runs
// form their own band — the guest view can then filter bands by
// visibility without any per-request recomputation.
$key = implode('|', [
$run->label ?? '-',
$run->model_id ?? '-',
$run->model_quant ?? '-',
$run->gpu ?? '-',
$run->gpu_count ?? '-',
$run->user_id ?? '-',
$run->visibility,
]);
$sessions[$key] ??= [
'key' => $key,
'label' => $run->label,
'model_id' => $run->model_id,
'model_quant' => $run->model_quant,
'gpu' => $run->gpu,
'gpu_count' => $run->gpu_count,
'runner_name' => $run->runner_name,
'user_id' => $run->user_id,
'visibility' => $run->visibility,
'ran_at' => $run->ran_at,
'passed' => 0,
'failed' => 0,
'tokens' => 0,
'runs' => collect(),
];
$agg = $this->aggregateRun($run);
$sessions[$key]['passed'] += $agg['passed'];
$sessions[$key]['failed'] += $agg['failed'];
$sessions[$key]['tokens'] += $agg['total_tokens'];
$sessions[$key]['runs']->push($agg + ['suite' => $run->suite, 'level' => $agg['level']]);
}
DB::transaction(function () use ($sessions) {
DB::table('session_summaries')->truncate();
foreach (array_values($sessions) as $bandIndex => $band) {
$summaryId = DB::table('session_summaries')->insertGetId([
'session_key' => $band['key'],
'label' => $band['label'],
'model_id' => $band['model_id'],
'model_quant' => $band['model_quant'],
'gpu' => $band['gpu'],
'gpu_count' => $band['gpu_count'],
'runner_name' => $band['runner_name'],
'user_id' => $band['user_id'],
'visibility' => $band['visibility'],
'ran_at' => $band['ran_at'],
'total_passed' => $band['passed'],
'total_failed' => $band['failed'],
'total_tokens' => $band['tokens'],
'sequence' => $bandIndex,
'created_at' => now(),
'updated_at' => now(),
]);
foreach ($band['runs'] as $suiteIndex => $row) {
DB::table('session_suite_rows')->insert([
'session_summary_id' => $summaryId,
'benchmark_run_id' => $row['run_pk'],
'run_id' => $row['run_id'],
'suite' => $row['suite'],
'level' => $row['level'],
'passed' => $row['passed'],
'failed' => $row['failed'],
'avg_user_tps' => $row['avg_user_tps'],
'weighted_user_tps' => $row['weighted_user_tps'],
'total_tokens' => $row['total_tokens'],
'sequence' => $suiteIndex,
'created_at' => now(),
'updated_at' => now(),
]);
}
}
});
$this->info('session summaries rebuilt: '.count($sessions).' session(s).');
return self::SUCCESS;
}
/**
* Aggregate one run's suite table: pass/fail counts, simple + weighted
* user tok/s, generated tokens, and the run's thinking level.
*/
private function aggregateRun(object $run): array
{
$model = Suites::modelFor($run->suite);
if ($model === null) {
return [
'run_pk' => $run->id,
'run_id' => $run->run_id,
'level' => null,
'passed' => 0,
'failed' => 0,
'avg_user_tps' => null,
'weighted_user_tps' => null,
'total_tokens' => 0,
];
}
$table = $model::make()->getTable();
if ($run->suite === 'scenariobench') {
$agg = DB::table($table)->where('benchmark_run_id', $run->id)
->selectRaw('count(*) filter (where ok) as passed')
->selectRaw('count(*) filter (where not ok) as failed')
->selectRaw('sum(gen_tokens) as total_tokens')
->first();
return [
'run_pk' => $run->id,
'run_id' => $run->run_id,
'level' => $this->runLevel($table, $run->id),
'passed' => (int) ($agg->passed ?? 0),
'failed' => (int) ($agg->failed ?? 0),
'avg_user_tps' => null,
'weighted_user_tps' => null,
'total_tokens' => (int) ($agg->total_tokens ?? 0),
];
}
$agg = DB::table($table)->where('benchmark_run_id', $run->id)
->selectRaw('count(*) filter (where ok) as passed')
->selectRaw('count(*) filter (where not ok) as failed')
->selectRaw('avg(user_tps) as avg_user_tps')
->selectRaw('sum(user_tps * total) filter (where total > 0) / nullif(sum(total) filter (where total > 0), 0) as weighted_user_tps')
->selectRaw('sum(total) as total_tokens')
->first();
return [
'run_pk' => $run->id,
'run_id' => $run->run_id,
'level' => $this->runLevel($table, $run->id),
'passed' => (int) ($agg->passed ?? 0),
'failed' => (int) ($agg->failed ?? 0),
'avg_user_tps' => $agg->avg_user_tps !== null ? round((float) $agg->avg_user_tps, 1) : null,
'weighted_user_tps' => $agg->weighted_user_tps !== null ? round((float) $agg->weighted_user_tps, 1) : null,
'total_tokens' => (int) ($agg->total_tokens ?? 0),
];
}
/**
* The run's thinking level: the distinct level across its result rows
* (a run is single-level; falls back to null when mixed or empty).
*/
private function runLevel(string $table, int $runPk): ?string
{
$level = DB::table($table)->where('benchmark_run_id', $runPk)
->distinct()->pluck('level')->filter()->values();
return $level->count() === 1 ? $level->first() : null;
}
}