|
|
<!DOCTYPE html>
|
|
|
<html lang="ko">
|
|
|
<head>
|
|
|
<meta charset="UTF-8">
|
|
|
<title>γκ°μλλ μμ±ν AI λͺ¨λΈ μ€μ¦ νκ° λ³΄κ³ μ</title>
|
|
|
<style>
|
|
|
@media print {
|
|
|
body { -webkit-print-color-adjust: exact; }
|
|
|
.no-print { display: none; }
|
|
|
.page-break { page-break-before: always; }
|
|
|
}
|
|
|
body { font-family: 'Malgun Gothic', sans-serif; margin: 0; padding: 0; background-color: #fff; }
|
|
|
.container { max-width: 800px; margin: 40px auto; padding: 40px; border: 1px solid #ccc; box-shadow: 0 0 10px rgba(0,0,0,0.1); }
|
|
|
h1, h2 { text-align: center; border-bottom: 2px solid #000; padding-bottom: 10px; margin-bottom: 40px; }
|
|
|
h1 { background: linear-gradient(135deg, #667eea 0%, #764ba2 100%); color: white; padding: 20px; border: none; }
|
|
|
h2 { border-top: 2px solid #ccc; margin-top: 40px; }
|
|
|
table { width: 100%; border-collapse: collapse; margin-bottom: 30px; }
|
|
|
th, td { border: 1px solid #999; padding: 10px; text-align: left; word-break: break-all; }
|
|
|
th { background-color: #f2f2f2; font-weight: bold; text-align: center; }
|
|
|
.label { font-weight: bold; color: #333; }
|
|
|
.score { font-size: 1.1em; color: #0056b3; font-weight: bold; }
|
|
|
.final-score { font-size: 1.2em; color: #d9534f; font-weight: bold; }
|
|
|
.footer { margin-top: 80px; text-align: center; }
|
|
|
.signature-box { margin-top: 60px; float: right; }
|
|
|
.button-container { text-align: center; margin-top: 40px; }
|
|
|
button { background-color: #007bff; color: white; border: none; padding: 10px 20px; font-size: 1em; cursor: pointer; border-radius: 5px; }
|
|
|
.total-score-row th, .total-score-row td { background-color: #e9ecef; font-weight: bold; font-size: 1.1em; }
|
|
|
.details-section h3 { color: #34495e; border-bottom: 1px solid #ccc; padding-bottom: 5px; }
|
|
|
.details-section pre { background-color: #f8f9fa; padding: 15px; border-radius: 5px; white-space: pre-wrap; word-wrap: break-word; font-size: 0.9em; }
|
|
|
</style>
|
|
|
</head>
|
|
|
<body>
|
|
|
|
|
|
<div class="container">
|
|
|
<h1>γκ°μλλ μμ±ν AI λͺ¨λΈ μ€μ¦ νκ° λ³΄κ³ μ</h1>
|
|
|
|
|
|
<table>
|
|
|
<tr>
|
|
|
<th style="width: 30%;">νκ° λμ LLM λͺ¨λΈ μ£Όμ</th>
|
|
|
<td colspan="3">{{ all_results.target_url if all_results.target_url else 'μ
λ ₯λμ§ μμ' }}</td>
|
|
|
</tr>
|
|
|
<tr>
|
|
|
<th>νκ° μμ</th>
|
|
|
<th>νκ° νλͺ©</th>
|
|
|
<th>κ²°κ³Όκ°</th>
|
|
|
<th>μ΅μ’
μ μ</th>
|
|
|
</tr>
|
|
|
|
|
|
<tr>
|
|
|
<th rowspan="3" style="vertical-align: middle;">μ λ νκ° (8μ )</th>
|
|
|
<td>1. Perplexity</td>
|
|
|
<td style="text-align: center;"><span class="score">{{ all_results.perplexity.score_display if all_results.perplexity and not all_results.perplexity.error else '-' }}</span></td>
|
|
|
<td style="text-align: center;"><span class="final-score">{{ "%.2f"|format(all_results.perplexity.final_score if all_results.perplexity and not all_results.perplexity.error else 0) }} / 3</span></td>
|
|
|
</tr>
|
|
|
<tr>
|
|
|
<td>2. ROUGE</td>
|
|
|
<td style="text-align: center;"><span class="score">{{ all_results.rouge.score_display if all_results.rouge and not all_results.rouge.error else '-' }}</span></td>
|
|
|
<td style="text-align: center;"><span class="final-score">{{ "%.2f"|format(all_results.rouge.final_score if all_results.rouge and not all_results.rouge.error else 0) }} / 3</span></td>
|
|
|
</tr>
|
|
|
<tr>
|
|
|
<td>3. BLEU</td>
|
|
|
<td style="text-align: center;"><span class="score">{{ all_results.bleu.score_display if all_results.bleu and not all_results.bleu.error else '-' }}</span></td>
|
|
|
<td style="text-align: center;"><span class="final-score">{{ "%.2f"|format(all_results.bleu.final_score if all_results.bleu and not all_results.bleu.error else 0) }} / 2</span></td>
|
|
|
</tr>
|
|
|
|
|
|
<tr>
|
|
|
<th rowspan="4" style="vertical-align: middle;">μ μ± νκ° (12μ )</th>
|
|
|
<td>4. MMLU</td>
|
|
|
<td style="text-align: center;"><span class="score">{{ all_results.mmlu.grade if all_results.mmlu and all_results.mmlu.grade else '-' }}</span></td>
|
|
|
<td style="text-align: center;"><span class="final-score">{{ "%.2f"|format(all_results.mmlu.final_score if all_results.mmlu and all_results.mmlu.final_score else 0) }} / 3</span></td>
|
|
|
</tr>
|
|
|
<tr>
|
|
|
<td>5. TruthfulQA</td>
|
|
|
<td style="text-align: center;"><span class="score">{{ all_results.truthfulqa.grade if all_results.truthfulqa and all_results.truthfulqa.grade else '-' }}</span></td>
|
|
|
<td style="text-align: center;"><span class="final-score">{{ "%.2f"|format(all_results.truthfulqa.final_score if all_results.truthfulqa and all_results.truthfulqa.final_score else 0) }} / 3</span></td>
|
|
|
</tr>
|
|
|
<tr>
|
|
|
<td>6. DROP</td>
|
|
|
<td style="text-align: center;"><span class="score">{{ all_results.drop.grade if all_results.drop and all_results.drop.grade else '-' }}</span></td>
|
|
|
<td style="text-align: center;"><span class="final-score">{{ "%.2f"|format(all_results.drop.final_score if all_results.drop and all_results.drop.final_score else 0) }} / 3</span></td>
|
|
|
</tr>
|
|
|
<tr>
|
|
|
<td>7. MBPP & HumanEval</td>
|
|
|
<td style="text-align: center;"><span class="score">{{ all_results.mbpp_humaneval.grade if all_results.mbpp_humaneval and all_results.mbpp_humaneval.grade else '-' }}</span></td>
|
|
|
<td style="text-align: center;"><span class="final-score">{{ "%.2f"|format(all_results.mbpp_humaneval.final_score if all_results.mbpp_humaneval and all_results.mbpp_humaneval.final_score else 0) }} / 3</span></td>
|
|
|
</tr>
|
|
|
|
|
|
{% set total_score = (all_results.perplexity.final_score|default(0, true) if not all_results.perplexity.error|default(false, true) else 0) +
|
|
|
(all_results.rouge.final_score|default(0, true) if not all_results.rouge.error|default(false, true) else 0) +
|
|
|
(all_results.bleu.final_score|default(0, true) if not all_results.bleu.error|default(false, true) else 0) +
|
|
|
(all_results.mmlu.final_score|default(0, true)) +
|
|
|
(all_results.truthfulqa.final_score|default(0, true)) +
|
|
|
(all_results.drop.final_score|default(0, true)) +
|
|
|
(all_results.mbpp_humaneval.final_score|default(0, true)) %}
|
|
|
<tr class="total-score-row">
|
|
|
<th colspan="3">μ΄μ </th>
|
|
|
<td style="text-align: center;"><span class="final-score">{{ "%.2f"|format(total_score) }} / 20</span></td>
|
|
|
</tr>
|
|
|
|
|
|
</table>
|
|
|
|
|
|
<div class="footer">
|
|
|
<p>μκΈ° λ΄μ©κ³Ό κ°μ΄ μμ±ν AI λͺ¨λΈ μ€μ¦ νκ°λ₯Ό μλ£νμμμ νμΈν©λλ€.</p>
|
|
|
<p class="eval-date"><b>νκ°μΌμ:</b> <span id="report-date"></span></p>
|
|
|
<div class="signature-box">
|
|
|
<p><b>νκ°μ:</b> ____________________ (μλͺ
)</p>
|
|
|
</div>
|
|
|
</div>
|
|
|
</div>
|
|
|
|
|
|
<div class="page-break"></div>
|
|
|
|
|
|
<div class="container details-section">
|
|
|
<h2>μμΈ νκ° λ΄μ©</h2>
|
|
|
|
|
|
<div class="detail-item">
|
|
|
<h3>1. Perplexity (μΈμ΄ λͺ¨λΈ νμ§)</h3>
|
|
|
<h4>μ
λ ₯ ν
μ€νΈ:</h4>
|
|
|
<pre>{{ input_texts.ppl_text if input_texts.ppl_text else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
{% if all_results.perplexity and all_results.perplexity.details %}
|
|
|
<h4>μΈλΆ μΈ‘μ κ°:</h4>
|
|
|
<pre>κΈ°λ³Έ PPL: {{ all_results.perplexity.details.base_ppl }}
|
|
|
νλν° κ³μ: {{ all_results.perplexity.details.penalty_factor }}
|
|
|
ν ν° μ: {{ all_results.perplexity.details.token_count }}
|
|
|
κ³μ° μκ°: {{ all_results.perplexity.details.calc_time }}</pre>
|
|
|
{% endif %}
|
|
|
</div>
|
|
|
|
|
|
<div class="detail-item">
|
|
|
<h3>2. ROUGE (μμ½ λ₯λ ₯)</h3>
|
|
|
<h4>λͺ¨λΈ μμ± μμ½λ¬Έ:</h4>
|
|
|
<pre>{{ input_texts.rouge_generated if input_texts.rouge_generated else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
<h4>μ°Έμ‘° μμ½λ¬Έ:</h4>
|
|
|
<pre>{{ input_texts.rouge_reference if input_texts.rouge_reference else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
{% if all_results.rouge and all_results.rouge.details %}
|
|
|
<h4>μΈλΆ μΈ‘μ κ°:</h4>
|
|
|
<pre>ROUGE-1: {{ all_results.rouge.details.rouge1 if all_results.rouge.details.rouge1 else '-' }}
|
|
|
ROUGE-2: {{ all_results.rouge.details.rouge2 if all_results.rouge.details.rouge2 else '-' }}
|
|
|
ROUGE-L: {{ all_results.rouge.details.rougeL if all_results.rouge.details.rougeL else '-' }}
|
|
|
κ°μ€ νκ· : {{ all_results.rouge.details.weighted_avg }}
|
|
|
κΈΈμ΄ νλν°: {{ all_results.rouge.details.length_penalty }}</pre>
|
|
|
{% endif %}
|
|
|
</div>
|
|
|
|
|
|
<div class="detail-item">
|
|
|
<h3>3. BLEU (λ²μ μ νμ±)</h3>
|
|
|
<h4>λͺ¨λΈ μμ± λ²μλ¬Έ:</h4>
|
|
|
<pre>{{ input_texts.bleu_generated if input_texts.bleu_generated else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
<h4>μ°Έμ‘° λ²μλ¬Έ 1:</h4>
|
|
|
<pre>{{ input_texts.bleu_reference1 if input_texts.bleu_reference1 else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
<h4>μ°Έμ‘° λ²μλ¬Έ 2:</h4>
|
|
|
<pre>{{ input_texts.bleu_reference2 if input_texts.bleu_reference2 else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
</div>
|
|
|
|
|
|
<div class="detail-item">
|
|
|
<h3>4. MMLU (λ³΅ν© λ¬Έμ ν΄κ²°)</h3>
|
|
|
<h4>λͺ¨λΈ μμ± κ²°κ³Ό:</h4>
|
|
|
<pre>{{ input_texts.mmlu_generated if input_texts.mmlu_generated else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
<h4>μ°Έμ‘°(μ λ΅):</h4>
|
|
|
<pre>{{ input_texts.mmlu_reference if input_texts.mmlu_reference else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
</div>
|
|
|
|
|
|
<div class="detail-item">
|
|
|
<h3>5. TruthfulQA (μ¬μ€ κΈ°λ° λ΅λ³)</h3>
|
|
|
<h4>λͺ¨λΈ μμ± κ²°κ³Ό:</h4>
|
|
|
<pre>{{ input_texts.truthfulqa_generated if input_texts.truthfulqa_generated else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
<h4>μ°Έμ‘°(μ λ΅):</h4>
|
|
|
<pre>{{ input_texts.truthfulqa_reference if input_texts.truthfulqa_reference else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
</div>
|
|
|
|
|
|
<div class="detail-item">
|
|
|
<h3>6. DROP (λ¬Έμ λ
ν΄/μΆλ‘ )</h3>
|
|
|
<h4>λͺ¨λΈ μμ± κ²°κ³Ό:</h4>
|
|
|
<pre>{{ input_texts.drop_generated if input_texts.drop_generated else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
<h4>μ°Έμ‘°(μ λ΅):</h4>
|
|
|
<pre>{{ input_texts.drop_reference if input_texts.drop_reference else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
</div>
|
|
|
|
|
|
<div class="detail-item">
|
|
|
<h3>7. MBPP & HumanEval (μ½λ μμ±/μ
무 μλν)</h3>
|
|
|
<h4>λͺ¨λΈ μμ± κ²°κ³Ό:</h4>
|
|
|
<pre>{{ input_texts.mbpp_humaneval_generated if input_texts.mbpp_humaneval_generated else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
<h4>μ°Έμ‘°(μ λ΅):</h4>
|
|
|
<pre>{{ input_texts.mbpp_humaneval_reference if input_texts.mbpp_humaneval_reference else '(νκ°λμ§ μμ)' }}</pre>
|
|
|
</div>
|
|
|
</div>
|
|
|
|
|
|
|
|
|
<div class="button-container no-print">
|
|
|
<button onclick="window.print()">λ³΄κ³ μ μΈμνκΈ°</button>
|
|
|
</div>
|
|
|
|
|
|
<script>
|
|
|
document.addEventListener('DOMContentLoaded', function() {
|
|
|
const now = new Date();
|
|
|
const year = now.getFullYear();
|
|
|
const month = String(now.getMonth() + 1).padStart(2, '0');
|
|
|
const day = String(now.getDate()).padStart(2, '0');
|
|
|
const hours = String(now.getHours()).padStart(2, '0');
|
|
|
const minutes = String(now.getMinutes()).padStart(2, '0');
|
|
|
const seconds = String(now.getSeconds()).padStart(2, '0');
|
|
|
const formattedDate = `${year}λ
${month}μ ${day}μΌ ${hours}:${minutes}:${seconds}`;
|
|
|
document.getElementById('report-date').textContent = formattedDate;
|
|
|
});
|
|
|
</script>
|
|
|
|
|
|
</body>
|
|
|
</html> |
