multipleof4/lynchmark
1
0
Fork 0
mirror of https://github.com/multipleof4/lynchmark.git synced 2026-01-13 16:17:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Feat: Calculate optimal temperature using weighted mean of correctness

Browse Source
This commit is contained in:
multipleof4
2025-11-18 12:01:10 -08:00
parent a3c0e76a85
commit 8aa2338b35
1 changed files with 8 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
gemini.html
View File

@@ -30,11 +30,13 @@
const genTimes=await fetch('./results.json').then(r=>r.json());
const models=readme.match(/<!-- GEMINI_START -->\n([\s\S]+?)\n<!-- MODELS_END -->/)[1].trim().split('\n').filter(Boolean);
const tests=[...new Set(Object.values(genTimes).flatMap(Object.keys))].sort();
const tStats={};
let [wSum, wTot] = [0, 0];
for(const model of models){
const sModel=model.replace(/[\/:]/g,'_');
const temp=model.split('TEMP:')[1]?.trim();
const tStr=model.split('TEMP:')[1];
const tVal=tStr?parseFloat(tStr):NaN;
const card=document.createElement('section');
card.className='rounded-2xl border border-gray-200 bg-white shadow-sm overflow-hidden';
card.innerHTML=`<div class="bg-gray-50 px-5 py-3 border-b border-gray-200"><p class="mono text-sm text-gray-700 font-medium">${model}</p></div><ul class="p-4 space-y-2" id="list-${sModel}"></ul>`;
@@ -67,18 +69,18 @@
li.innerHTML=`${status} <span class="font-medium text-gray-800">${test}</span><span class="mono text-gray-500 ml-auto">${time.toFixed(3)}s</span>`;
}
const ratio=tests.length?passed/tests.length:0;
if(temp){tStats[temp]??={p:0,t:0};tStats[temp].p+=passed;tStats[temp].t+=tests.length}
if(!isNaN(tVal)){wSum+=tVal*ratio;wTot+=ratio;}
const grade=gradeOf(ratio);
const scoreLi=document.createElement('li');
scoreLi.className='mt-3 pt-3 border-t border-gray-200 flex items-center text-sm justify-between';
scoreLi.innerHTML=`<span class="text-gray-600">Score</span><span class="flex items-center gap-3"><span class="mono text-gray-900 font-semibold">${passed}/${tests.length}</span><span class="inline-flex items-center rounded-full bg-blue-100 px-2 py-0.5 text-xs font-semibold text-blue-800">${grade}</span></span>`;
list.appendChild(scoreLi);
}
if(Object.keys(tStats).length){
const best=Object.entries(tStats).sort(([,a],[,b])=>b.p/b.t-a.p/a.t)[0];
if(wTot>0){
const optimal=(wSum/wTot).toFixed(4);
const statDiv=document.createElement('div');
statDiv.className='mt-8 p-6 bg-white rounded-2xl border border-gray-200 shadow-sm text-center';
statDiv.innerHTML=`<h3 class="text-gray-500 font-medium text-sm uppercase tracking-wider mb-2">Best Temperature</h3><div class="text-3xl font-bold text-gray-900">${best[0]}</div><div class="text-sm text-gray-500 mt-1">Pass Rate: ${((best[1].p/best[1].t)*100).toFixed(1)}%</div>`;
statDiv.innerHTML=`<h3 class="text-gray-500 font-medium text-sm uppercase tracking-wider mb-2">Optimal Temperature</h3><div class="text-3xl font-bold text-gray-900">${optimal}</div><div class="text-xs text-gray-400 mt-2">Weighted Center of Correctness</div>`;
container.appendChild(statDiv);
}
})();