File size: 6,676 Bytes
8eecd10
 
384e89d
8eecd10
 
 
 
384e89d
8eecd10
 
 
384e89d
8eecd10
 
 
 
 
 
 
 
 
 
384e89d
 
 
 
 
 
 
 
 
 
8eecd10
 
 
 
384e89d
 
8eecd10
 
384e89d
 
 
8eecd10
384e89d
8eecd10
 
 
 
 
 
 
384e89d
8eecd10
384e89d
 
8eecd10
384e89d
 
 
 
 
 
 
 
 
 
 
 
8eecd10
 
 
 
 
384e89d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8eecd10
 
 
384e89d
8eecd10
 
 
 
 
384e89d
8eecd10
 
 
 
 
 
 
 
 
 
384e89d
 
8eecd10
 
 
 
 
 
 
 
 
 
 
 
384e89d
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
<!DOCTYPE html>
<html lang="en">

<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>Zhiyin</title>
    <link rel="stylesheet" href="styles.css?v=3">
    <link href="https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700&display=swap" rel="stylesheet">
    <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.0.0/css/all.min.css">
</head>

<body>
    <div class="container">
        <header class="header">
            <h1><i class="fas fa-feather"></i> Zhiyin</h1>
            <p>Exploring the Frontier of Chinese LLM Writing</p>
        </header>

        <div class="dashboard card">
            <section class="overview-section">
                <h2 class="section-title">
                    <span class="accent-bar"></span>
                    Benchmark Overview
                    <span class="section-title-spacer"></span>
                    <span class="external-links-text">
                        <a href="https://github.com/zake7749/Chinese-Writing-Bench" target="_blank" rel="noopener"
                            class="external-link-text">GitHub</a>
                        <span class="divider">&bull;</span>
                        <a href="https://huggingface.co/datasets/zake7749/chinese-writing-benchmark" target="_blank"
                            rel="noopener" class="external-link-text">Hugging Face</a>
                    </span>
                </h2>
                <div class="overview-content">
                    <div class="overview-text">
                        <p>
                            <strong>Zhiyin</strong> is an LLM-as-a-judge benchmark for Chinese writing evaluation,
                            featuring 280 test cases across 18 diverse writing tasks in this V1 release.
                        </p>
                        <p>
                            Our method relies on <strong>pairwise comparison</strong>. A powerful language model acts as
                            the judge, scoring a model's response relative to a fixed baseline (GPT-4.1), which is
                            anchored at a score of 5.
                        </p>

                        <h4>Scoring System</h4>
                        <p>The judge assigns the model's response an integer score from 0 to 10, where:</p>
                        <ul>
                            <li>A score > 5 indicates the response is <strong>superior</strong> to the baseline.</li>
                            <li>A score = 5 indicates the response is <strong>on par</strong> with the baseline.</li>
                            <li>A score < 5 indicates the response is <strong>inferior</strong> to the baseline.</li>
                        </ul>

                        <h4>Evaluation Dimensions</h4>
                        <p>To ensure a comprehensive analysis, the final score is informed by a multi-dimensional
                            assessment. The judge evaluates the response across six key criteria:</p>
                        <ol>
                            <li><strong>Comprehension & Relevance:</strong> How well the response understands the
                                prompt's intent and stays on topic.</li>
                            <li><strong>Structure & Coherence:</strong> How clear, logical, and well-organized the
                                writing is.</li>
                            <li><strong>Prose & Style:</strong> The quality of the language, grammar, and adherence to
                                the requested tone.</li>
                            <li><strong>Creativity & Originality:</strong> The novelty of the ideas and the uniqueness
                                of the perspective.</li>
                            <li><strong>Depth & Insight:</strong> The level of detail, analysis, and substance provided.
                            </li>
                            <li><strong>Helpfulness:</strong> How effectively the response fulfills the user's overall
                                goal.</li>
                        </ol>
                    </div>
                </div>
            </section>

            <section class="dashboard-controls">
                <div class="controls-header">
                    <div class="judge-selector">
                        <span class="control-label">Judge Model</span>
                        <div class="judge-toggle">
                            <button class="judge-btn active" data-judge="gpt5.4">GPT-5.4</button>
                            <button class="judge-btn" data-judge="o3">O3</button>
                        </div>
                    </div>
                    <div class="search-container">
                        <i class="fas fa-search search-icon"></i>
                        <input type="text" id="globalSearch" class="search-input" placeholder="Search models...">
                    </div>
                </div>

                <div class="tabs-container">
                    <div class="tabs-header">
                        <button class="tab-btn active" data-target="generalTableSection">All Writing Tasks</button>
                        <button class="tab-btn" data-target="complicatedTableSection">Complicated Writing Tasks</button>
                    </div>
                </div>
            </section>

            <section class="tab-content active" id="generalTableSection">
                <div id="generalTable" class="table-container">
                    <!-- General Writing table will be populated here -->
                </div>
            </section>

            <section class="tab-content" id="complicatedTableSection">
                <div id="complicatedTable" class="table-container">
                    <!-- Complicated Writing table will be populated here -->
                </div>
            </section>

            <section class="citation-section">
                <h2 class="section-title"><span class="accent-bar"></span>Citation</h2>
                <div class="citation-content">
                    <p>
                        If you use these results, please cite our paper:<br>
                        <em>"Zhiyin: Exploring the Frontier of Chinese LLM Writing, 2025.
                            https://github.com/zake7749/Chinese-Writing-Bench"</em>
                    </p>
                </div>
            </section>
        </div>

        <div id="loading" class="loading">
            <div class="spinner"></div>
            <p>Loading benchmark data...</p>
        </div>
    </div>
    <script src="script.js"></script>
</body>

</html>