token-usage.js

'use strict';
const TokenUsage={
  aggregate(answers){
    const t={responses:0,input_total:0,output_total:0,input_avg:null,output_avg:null};
    for(const a of answers){const u=a?.usage||{},i=u.prompt_tokens??u.input_tokens,o=u.completion_tokens??u.output_tokens;if(!Number.isInteger(i)||!Number.isInteger(o)||i<0||o<0)continue;t.responses++;t.input_total+=i;t.output_total+=o;}
    if(t.responses){t.input_avg=t.input_total/t.responses;t.output_avg=t.output_total/t.responses;}return t;
  },
  format(n,avg=false){return typeof n==='number'&&Number.isFinite(n)?n.toLocaleString('en-US',{maximumFractionDigits:avg?1:0}):'—';},
  table(d,condition){
    const e=s=>String(s??'').replace(/[&<>"']/g,c=>({'&':'&amp;','<':'&lt;','>':'&gt;','"':'&quot;',"'":'&#39;'}[c]));
    return `<div class="table-wrap token-summary"><table><caption>성능 · 응답 생성 토큰</caption><thead><tr><th>Model</th><th>응답 / 채점</th><th>정확도</th><th>총 입력 토큰</th><th>문항당 평균 입력</th><th>총 출력 토큰</th><th>문항당 평균 출력</th><th>토큰 집계 문항</th></tr></thead><tbody>${d.settings.filter(s=>s.condition===condition).map(s=>{const t=s.tokens,known=t&&t.responses>0;return `<tr><th>${e(s.label)}</th><td>${s.read} / ${s.graded}</td><td>${CleanResults.cell({n:d.total,graded:s.graded,correct:s.correct})}</td><td>${known?this.format(t.input_total):'—'}</td><td>${known?this.format(t.input_avg,true):'—'}</td><td>${known?this.format(t.output_total):'—'}</td><td>${known?this.format(t.output_avg,true):'—'}</td><td>${t?`${t.responses} / ${s.read}`:'집계 대기'}</td></tr>`;}).join('')}</tbody></table></div><p class="caption">완료된 응답 중 토큰 사용량이 기록된 문항 기준 · 채점/임베딩/실패 재시도 제외 · 입력에는 보고된 캐시 토큰 포함 · 모델별 토크나이저가 달라 동일 자료도 토큰 수가 다릅니다.</p>`;
  },
  answer(a){const t=this.aggregate([a]);return t.responses?`입력 ${this.format(t.input_total)} · 출력 ${this.format(t.output_total)} 토큰`:'토큰 사용량 미기록';}
};