app.py

import gradio as gr
from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
import pandas as pd

# 🎨 增强后的自定义 CSS
custom_css = """
/* 全局设置：简洁、高级的字体和背景 */
:root {
    --color-background-primary: #f8f8f8; /* 浅米白色背景 */
    --color-background-secondary: #ffffff; /* 卡片背景 */
    --color-text-primary: #333333;
    --color-accent: #8e80ff; /* 浅紫色强调色 (Primary) */
    --color-accent-light: #a99dff; /* 浅紫色悬停色 */
    --shadow-medium: 0 4px 12px rgba(0, 0, 0, 0.08);
}

body {
    background-color: var(--color-background-primary) !important;
}

/* 增加容器最大宽度以展示完整表格 */
.gradio-container {
    max-width: 1400px; /* 宽度从 1800px 调窄到 1400px */
    margin: 0 auto;
    padding: 20px;
}

/* 标题样式 */
#space-title {
    color: var(--color-text-primary);
    font-size: 3em;
    font-weight: 700;
    margin-bottom: 0.5em;
    padding-top: 20px;
}

/* Group/Block 组件的卡片样式 */
.gr-group, .gr-block {
    background-color: var(--color-background-secondary);
    border-radius: 12px;
    box-shadow: var(--shadow-medium); 
    transition: box-shadow 0.3s ease;
    padding: 15px;
    margin-bottom: 20px;
}

.gr-group:hover, .gr-block:hover {
    box-shadow: 0 6px 18px rgba(0, 0, 0, 0.12); 
}

/* Leaderboard 容器：调整内部布局的关键 */
.leaderboard_root {
    padding: 0 !important; 
}

/* 搜索栏布局调整 (第一行) */
.leaderboard_root > div:nth-child(1) {
    padding: 0 15px 15px 15px;
}

/* 过滤器和列选择布局调整 (第二行) */
.leaderboard_root > div:nth-child(2) {
    display: flex; 
    padding: 0 15px 15px 15px;
}

.leaderboard_root .gr-form {
    border: none;
}

/* Search Bar */
#search-bar-table-box {
    width: 100%;
    margin-bottom: 10px;
}
#search-bar-table-box > div:first-child {
    background: none;
    border: none;
}

/* === Select Columns to Display: 强制单行展示 === */
/* 定位 SelectColumns 的内部复选框容器 */
.leaderboard-filter-column:first-child .gr-form-checkbox-group {
    /* 使用 flex 容器 */
    display: flex !important;
    flex-wrap: nowrap !important; /* 强制不换行 */
    overflow-x: auto !important; /* 允许水平滚动 */
    gap: 10px; 
    padding-bottom: 5px; 
}

/* 确保每个复选框标签保持内联块级元素 */
.leaderboard-filter-column:first-child .gr-form-checkbox-group label {
    flex-shrink: 0 !important; /* 防止选项被压缩 */
    display: inline-block !important; /* 确保每个选项占据其自然宽度 */
    margin: 0;
    white-space: nowrap; /* 确保文字也不换行 */
}

#leaderboard-table, #leaderboard-table-lite {
    margin-top: 15px;
    border-radius: 8px;
    overflow: hidden; 
}

#leaderboard-table th {
    background-color: var(--color-accent);
    color: white;
    font-weight: 600;
    text-transform: uppercase;
    border-bottom: 2px solid var(--color-accent-light);
}

#leaderboard-table tr:hover {
    background-color: #f0f0f0; 
    cursor: pointer;
    transition: background-color 0.2s ease;
}

#leaderboard-table td:nth-child(2),
#leaderboard-table th:nth-child(2) {
    max-width: 400px;
    overflow: auto;
    white-space: nowrap;
}

#leaderboard-table td:nth-child(3) {
    font-weight: bold;
    color: var(--color-accent);
}

/* Citation 区域 */
#citation-group {
    padding: 20px; 
    margin-top: 10px;
}

#citation-button {
    margin-top: 0;
    padding: 0;
}

/* 修复 Citation 复制图标重叠问题 */
#citation-button label {
    display: block; 
    position: relative;
}

#citation-button textarea {
    font-family: monospace;
    background-color: #f1f1f1;
    border: 1px solid #cccccc;
    border-radius: 6px;
    padding: 10px;
    padding-right: 40px !important; /* 为复制按钮腾出空间 */
    font-size: 14px !important;
    width: 100% !important; 
    box-sizing: border-box; 
}

/* 调整复制按钮的位置 */
#citation-button > label > button {
    position: absolute;
    top: 10px; 
    right: 10px; 
    margin: 0;
    transform: scale(1.1); 
    transition: transform 0.2s ease;
    background-color: var(--color-accent) !important;
    color: white !important;
    border: none !important;
    border-radius: 6px;
    z-index: 10; 
}

#citation-button > label > button:hover {
    transform: scale(1.2); 
    background-color: var(--color-accent-light) !important;
}

/* Leaderboard 内部过滤/选择组件微调 */
.leaderboard_root .leaderboard-filter-column:last-child {
    flex-grow: 1; 
    max-width: 50%;
}

.leaderboard_root .leaderboard-filter-column:first-child {
    max-width: 50%; 
    padding-right: 20px;
}

#filter_type{
    border: 0;
    padding-left: 0;
    padding-top: 0;
}
#filter_type label {
    display: flex;
}
#filter_type label > span{
    margin-top: var(--spacing-lg);
    margin-right: 0.5em;
}
#filter_type label > .wrap{
    width: 103px;
}
#filter_type label > .wrap .wrap-inner{  
    padding: 2px;
}
#filter_type label > .wrap .wrap-inner input{
    width: 1px
}
#filter-columns-type{
    border:0;
    padding:0.5;
}
#filter-columns-size{
    border:0;
    padding:0.5;
}
#box-filter > .form{
    border: 0
}
/* 其他 Gradio 元素的简洁化 */
.wrap-inner input[type="text"], .wrap-inner input[type="number"] {
    border-radius: 6px;
    border: 1px solid #cccccc;
    padding: 8px 12px;
}
"""

get_window_url_params = """
    function(url_params) {
        const params = new URLSearchParams(window.location.search);
        url_params = Object.fromEntries(params);
        return url_params;
    }
    """


TITLE = """<h1 align="center" id="space-title">SGI-Bench Leaderboard 🏆</h1>"""
INFO = """<p align="center">
    <a href="https://internscience.github.io/SGI-Page/"><b>🌐Official Site</b></a> ·
    <a href="https://arxiv.org/pdf/2512.16969"><b>📜arXiv</b></a> ·
    <a href="https://huggingface.co/collections/InternScience/sgi-bench"><b>🤗Hugging Face</b></a> ·
    <a href="https://github.com/InternScience/SGI-Bench"><b>💻GitHub</b></a>
</p>"""

CITATION_BUTTON_LABEL = "📖 Citation"
CITATION_BUTTON_TEXT = r"""@article{xu2025probing,
  title={Probing Scientific General Intelligence of LLMs with Scientist-Aligned Workflows},
  author={Xu, Wanghan and Zhou, Yuhao and Zhou, Yifan and Cao, Qinglong and Li, Shuo and Bu, Jia and Liu, Bo and Chen, Yixin and He, Xuming and Zhao, Xiangyu and others},
  journal={arXiv preprint arXiv:2512.16969},
  year={2025}
}"""

LEADERBOARD_DATA = [
    {"name": "Intern-S1", "type": "Open", "scores": [15.74, 38.09, 28.79, 29.02, 28.87]},
    {"name": "Intern-S1-mini", "type": "Open", "scores": [11.06, 36.04, 16.97, 12.42, 16.84]},
    {"name": "Qwen3-VL-235B-A22B", "type": "Open", "scores": [11.97, 39.28, 28.41, 30.30, 31.62]},
    {"name": "Qwen3-8B", "type": "Open", "scores": [8.18, 35.78, 18.45, 9.96, 23.37]},
    {"name": "Llama-4-Scout", "type": "Open", "scores": [7.86, 29.72, 20.37, 21.66, 25.77]},
    {"name": "Qwen3-Max", "type": "Closed", "scores": [15.38, 39.83, 33.21, 33.62, 37.80]},
    {"name": "GPT-4o", "type": "Closed", "scores": [7.86, 35.95, 26.94, 31.31, 32.30]},
    {"name": "GPT-4.1", "type": "Closed", "scores": [11.32, 36.49, 34.32, 36.63, 38.49]},
    {"name": "GPT-5", "type": "Closed", "scores": [14.47, 55.40, 29.89, 16.31, 38.14]},
    {"name": "GPT-5.1", "type": "Closed", "scores": [11.64, 47.12, 31.00, 22.77, 34.02]},
    {"name": "GPT-5.2-Pro", "type": "Closed", "scores": [15.72, 55.03, 28.04, 17.50, 39.18, 31.09]},
    {"name": "o3", "type": "Closed", "scores": [12.89, 46.07, 31.73, 30.04, 32.65]},
    {"name": "o4-mini", "type": "Closed", "scores": [11.95, 40.78, 35.79, 28.86, 33.33]},
    {"name": "Gemini-2.5-Flash", "type": "Closed", "scores": [10.69, 39.13, 21.03, 18.55, 34.36]},
    {"name": "Gemini-2.5-Pro", "type": "Closed", "scores": [15.09, 39.95, 22.51, 22.05, 41.24]},
    {"name": "Gemini-3-Pro", "type": "Closed", "scores": [18.48, 39.68, 36.64, 32.45, 41.92]},
    {"name": "Claude-Opus-4.1", "type": "Closed", "scores": [12.93, 40.29, 34.69, 25.38, 38.83]},
    {"name": "Claude-Sonnet-4.5", "type": "Closed", "scores": [13.84, 43.20, 35.79, 30.15, 37.80]},
    {"name": "Grok-4", "type": "Closed", "scores": [13.31, 37.12, 33.71, 29.01, 30.24]},
]

def build_leaderboard_df():
    task_cols = ["Scientific Deep Research", "Idea Generation", "Dry Experiment", "Wet Experiment", "Experimental Reasoning"]
    rows = []
    for item in LEADERBOARD_DATA:
        name = item["name"]
        type = item["type"]
        scores = item["scores"]
        row = {
            "Model": name,
            "Type": type,
            "SGI-Score": round(sum(scores) / len(scores), 2),
        }
        for i, col in enumerate(task_cols):
            row[col] = scores[i]
        rows.append(row)
    cols = ["Model", "Type", "SGI-Score"] + task_cols
    df = pd.DataFrame(rows, columns=cols).sort_values(by=["SGI-Score"], ascending=False).round(decimals=2)
    return df

LEADERBOARD_DF = build_leaderboard_df()


def init_leaderboard(dataframe):
    datatypes = ["str", "str", "number", "number", "number", "number", "number", "number"]
    default_selection = ["Model","Type","SGI-Score","Scientific Deep Research","Idea Generation","Dry Experiment","Wet Experiment","Experimental Reasoning"]
    cant_deselect = ["Model","Type",]
    return Leaderboard(
        value=dataframe,
        datatype=datatypes,
        select_columns=SelectColumns(
            default_selection=default_selection,
            cant_deselect=cant_deselect,
            label="Select Columns to Display:",
        ),
        search_columns=["Model"],
        hide_columns=[],
        filter_columns=[ColumnFilter("Type", type="checkboxgroup", label="Model Types:")],
        interactive=False,
    )


demo = gr.Blocks(css=custom_css, theme=gr.themes.Soft())
with demo:
    gr.HTML(TITLE)
    gr.HTML(INFO)
    
    # Leaderboard 主体包含 Search, Filters, Select Columns 和 Table
    with gr.Group(elem_id="leaderboard-group"):
        leaderboard = init_leaderboard(LEADERBOARD_DF)

    with gr.Row():
        with gr.Column():
            with gr.Group(elem_id="citation-group"):
                citation_button = gr.Textbox(
                    value=CITATION_BUTTON_TEXT,
                    label=CITATION_BUTTON_LABEL,
                    lines=CITATION_BUTTON_TEXT.count('\n')+1,
                    elem_id="citation-button",
                    show_copy_button=True,
                    interactive=False 
                )

demo.queue(default_concurrency_limit=40).launch()