remove leaderboard, replace by gradio components

app.py

@@ -2,7 +2,6 @@ import gradio as gr
 import gradio.components as grc
 import pandas as pd
 from apscheduler.schedulers.background import BackgroundScheduler
-from gradio_leaderboard import ColumnFilter, Leaderboard, SelectColumns
 from huggingface_hub import snapshot_download
 from rich import print
 
@@ -78,44 +77,178 @@ LEADERBOARD_DF = get_leaderboard_df(
 ) = get_evaluation_queue_df(settings.EVAL_REQUESTS_PATH, EVAL_COLS)
 
 
-def init_leaderboard(dataframe: pd.DataFrame, cols: list[str]) -> Leaderboard:
-    if dataframe is None or dataframe.empty:
-        raise ValueError("Leaderboard DataFrame is empty or None.")
-    # print("///// --- dataframe.head() --- /////", Markdown(dataframe.head().to_markdown() or "No data"))
-    selected_columns = SelectColumns(
-        default_selection=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and c.name in cols],
-        cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden and c.name in cols],
-        label="Select Columns to Display:",
+def filter_dataframe_by_columns(selected_cols: list[str], original_df: pd.DataFrame) -> pd.DataFrame:
+    """
+    根据选择的列过滤 DataFrame
+    """
+    # 始终包含基础列 'T' 和 'Model'
+    base_cols = ['T', 'Model']
+    all_selected_cols = [col for col in base_cols if col in original_df.columns]
+    
+    # 添加用户选择的列（排除已存在的基础列）
+    for col in selected_cols:
+        if col in original_df.columns and col not in all_selected_cols:
+            all_selected_cols.append(col)
+    
+    # 确保列的顺序：基础列在前，然后是按原始顺序的选中列
+    ordered_cols = []
+    for col in original_df.columns:
+        if col in all_selected_cols:
+            ordered_cols.append(col)
+    
+    # 确保总是返回 DataFrame，即使是单列也使用 [[]] 来保持 DataFrame 类型
+    if ordered_cols:
+        filtered_df = original_df.loc[:, ordered_cols]
+    else:
+        filtered_df = original_df
+    return filtered_df
+
+
+def filter_dataframe_by_precision(selected_precisions: list[str], df: pd.DataFrame) -> pd.DataFrame:
+    """
+    根据选择的 precision 筛选 DataFrame
+    如果没有选择 precision，返回空的 DataFrame
+    """
+    if not selected_precisions:
+        return df.iloc[0:0].copy()  # 返回相同结构但为空的 DataFrame
+    
+    precision_col = AutoEvalColumn.precision.name
+    if precision_col not in df.columns:
+        return df
+    
+    # 筛选包含任一选定 precision 的行
+    mask = df[precision_col].isin(selected_precisions)
+    filtered_df = df.loc[mask, :]
+    return filtered_df
+
+
+def search_models_in_dataframe(search_text: str, df: pd.DataFrame) -> pd.DataFrame:
+    """
+    在 DataFrame 中搜索包含关键词的 Model 名称
+    支持逗号分隔的多个关键词，匹配包含任一关键词的行
+    """
+    if not search_text or not search_text.strip():
+        return df
+    
+    # 分割逗号，去除空白并转换为小写用于匹配
+    import re
+    
+    keywords = [keyword.strip().lower() for keyword in search_text.split(',') if keyword.strip()]
+    if not keywords:
+        return df
+    
+    if 'Model' not in df.columns:
+        return df
+    
+    # 匹配函数：从 HTML 中提取纯文本并检查是否包含关键词
+    def matches_search(model_cell):
+        if pd.isna(model_cell):
+            return False
+        
+        # 从 HTML 链接中提取纯文本（model_name）
+        # 格式: <a ...>model_name</a> 或直接是文本
+        text = str(model_cell)
+        
+        # 提取 HTML 标签内的文本
+        # 匹配 <a>...</a> 标签内的内容，或直接使用文本
+        match = re.search(r'<a[^>]*>([^<]+)</a>', text, re.IGNORECASE)
+        if match:
+            model_name = match.group(1).lower()
+        else:
+            model_name = text.lower()
+        
+        # 检查是否包含任一关键词
+        return any(keyword in model_name for keyword in keywords)
+    
+    # 应用搜索过滤
+    mask = df['Model'].apply(matches_search)
+    filtered_df = df.loc[mask, :]
+    return filtered_df
+
+
+def init_leaderboard_tabs(dataframe: pd.DataFrame, cols: list[str]):
+    # 存储原始 DataFrame 以便后续过滤使用（使用闭包保存）
+    original_df = dataframe.copy()
+    
+    available_precisions = sorted(original_df["Precision"].dropna().unique().tolist())
+    default_precision = ['bfloat16'] if 'bfloat16' in available_precisions else (available_precisions[:1] if available_precisions else [])
+    
+    # 初始化显示的列（包含基础列和默认选中的列）
+    default_selected = [col for col in dataframe.columns if col in cols] + ['Average ⬆️']
+    
+    # 先按 precision 筛选 original_df
+    precision_filtered_df = filter_dataframe_by_precision(default_precision, original_df)
+    # 根据默认选择再筛选一次 DataFrame
+    initial_filtered_df = filter_dataframe_by_columns(default_selected, precision_filtered_df)
+    
+    with gr.Row():
+        with gr.Column(scale=1):
+            search = gr.Textbox(label="Search", placeholder="Separate multiple queries with commas")
+            show_columns = gr.CheckboxGroup(
+                choices=[col for col in dataframe.columns if col not in ['T', 'Model']],
+                label="Select Columns to Display",
+                value=default_selected,
+                interactive=True,
+            )
+        with gr.Column(scale=1):
+            model_type = gr.CheckboxGroup(
+                [],
+                label="Model Type",
+                value=[],
+            )
+            precision = gr.CheckboxGroup(
+                choices=available_precisions,
+                label="Precision",
+                value=default_precision,
+                interactive=True,
+            )
+            hide_models = gr.CheckboxGroup(
+                ['Deleted/incomplete'],
+                label="Hide Models",
+                value=['Deleted/incomplete'],
+                interactive=True,
+            )
+
+    with gr.Row():
+        with gr.Column(scale=3):
+            leaderboard = gr.Dataframe(
+                value=initial_filtered_df,  # 使用初始筛选后的 DataFrame
+                interactive=False,
+                wrap=False,
+                datatype='markdown',
+                elem_id="auto-width-dataframe",
+            )
+    
+    # 统一的更新函数：同时处理 precision、列筛选和搜索
+    def update_dataframe(search_text: str, selected_cols: list[str], selected_precisions: list[str]):
+        # 先按 precision 筛选 original_df
+        precision_filtered_df = filter_dataframe_by_precision(selected_precisions, original_df)
+        # 再按列筛选
+        column_filtered_df = filter_dataframe_by_columns(selected_cols, precision_filtered_df)
+        # 最后按搜索关键词筛选
+        final_df = search_models_in_dataframe(search_text, column_filtered_df)
+        return final_df
+    
+    # 绑定搜索、列选择和 precision 的变化事件，动态更新 DataFrame
+    search.change(
+        fn=update_dataframe,
+        inputs=[search, show_columns, precision],
+        outputs=leaderboard,
+    )
+    
+    show_columns.change(
+        fn=update_dataframe,
+        inputs=[search, show_columns, precision],
+        outputs=leaderboard,
     )
-    search_columns = [AutoEvalColumn.model.name, AutoEvalColumn.license.name]
-    hidden_columns = [c.name for c in fields(AutoEvalColumn) if c.hidden and c.name in cols]
-    filter_columns = [
-        ColumnFilter(AutoEvalColumn.model_type.name, type="checkboxgroup", label="Model types"),
-        ColumnFilter(AutoEvalColumn.precision.name, type="checkboxgroup", label="Precision"),
-        ColumnFilter(
-            AutoEvalColumn.params.name,
-            type="slider",
-            min=0.01,
-            max=150,
-            label="Select the number of parameters (B)",
-        ),
-        ColumnFilter(
-            AutoEvalColumn.still_on_hub.name,
-            type="boolean",  # pyright: ignore[reportArgumentType]
-            label="Deleted/incomplete",
-            default=False,
-        ),
-    ]
-    return Leaderboard(
-        value=dataframe,
-        datatype=[c.type for c in fields(AutoEvalColumn) if c.name in cols],
-        select_columns=selected_columns,
-        search_columns=search_columns,
-        hide_columns=hidden_columns,
-        filter_columns=filter_columns,  # pyright: ignore[reportArgumentType]
-        bool_checkboxgroup_label="Hide models",
-        interactive=False,
+    
+    precision.change(
+        fn=update_dataframe,
+        inputs=[search, show_columns, precision],
+        outputs=leaderboard,
     )
+    
+    return leaderboard
 
 
 demo = gr.Blocks(css_paths=[custom_css])
@@ -135,12 +268,12 @@ with demo:
                     cols,
                     benchmark_cols,
                 )
-                leaderboard = init_leaderboard(BENCHMARK_DF, cols)
+                leaderboard = init_leaderboard_tabs(BENCHMARK_DF, benchmark_cols)
 
-        with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=len(BENCHMARKS)):
+        with gr.TabItem("📝 About", elem_id="about-tab", id=len(BENCHMARKS)):
             gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
 
-        with gr.TabItem("🚀 Submit here! ", elem_id="llm-benchmark-tab-table", id=len(BENCHMARKS) + 1):
+        with gr.TabItem("🚀 Submit here! ", elem_id="submit-tab", id=len(BENCHMARKS) + 1):
             with gr.Column():
                 with gr.Row():
                     gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")

src/assets/css/custom.css

@@ -92,3 +92,16 @@
 #box-filter > .form{
     border: 0
 }
+
+/* Auto-width DataFrame: single line per column, width based on content */
+#auto-width-dataframe table {
+    table-layout: auto !important;
+    width: auto !important;
+}
+
+#auto-width-dataframe td,
+#auto-width-dataframe th {
+    white-space: nowrap !important;
+    overflow: hidden !important;
+    text-overflow: ellipsis !important;
+}