app.py

import pandas as pd
import gradio as gr
import os
import html
import requests
from dotenv import load_dotenv
from matplotlib.colors import LinearSegmentedColormap
import plotly.graph_objects as go
import numpy as np
from scipy.optimize import curve_fit
from huggingface_hub import HfApi
from huggingface_hub.utils import GatedRepoError, HfHubHTTPError
from gradio_rangeslider import RangeSlider
import datetime
from title import css, TITLE_HTML, SUBTITLE_HTML, LINKS_HTML
from data_manager import DataManager, LongContextDataManager
from longctx_utils import *

load_dotenv()
webhook_url = os.environ.get("WEBHOOK_URL")
metric_list = [
    "Compression Ratio (%)",
    "Bits Per Character (BPC)",
    "Bits Per Byte (BPB)",
]
model_size_list = [
    ">20B",
    "~14B",
    # "~9B",
    "~7B",
    "~3B",
    "~1.5B",
    "Other",
]
metric_to_sheet = {
    "Compression Ratio (%)": "cr",
    "Bits Per Character (BPC)": "bpc",
    "Bits Per Byte (BPB)": "bpb",
}
model_size_to_file_name = {
    ">20B": "20b+",
    "~14B": "14b",
    # "~9B": "9b",
    "~7B": "7b",
    "~3B": "3b",
    "~1.5B": "1b5",
    "Other": "other",
}
SCALING_EXTRAPOLATE_MAX_B = 10000
SCALING_FIT_POINTS = 200
SCALING_PLOT_WIDTH = 1000
SCALING_PLOT_HEIGHT = 620
SCALING_PLOT_MARGIN = dict(l=70, r=35, t=70, b=65)
MODEL_NAME_DISPLAY_MAX_CHARS = 28
FRONTIER_TABLE_COLUMNS = ["params", "model", "ratio%", "vs fit%", "row_bg", "row_hover_bg"]
FIT_RESIDUAL_GOOD_COLOR = "#2CA25F"
FIT_RESIDUAL_NEUTRAL_COLOR = "#F7F7F7"
FIT_RESIDUAL_BAD_COLOR = "#DE2D26"
FIT_LINE_COLOR = "#4B5563"


def format_fit_equation(a, b, c):
    if abs(c) < 0.01:
        return f"y = {a:.2f} × x^{b:.3f}"
    return f"y = {a:.2f} × x^{b:.3f} + {c:.2f}"


def build_fit_line_hovertemplate(label, a, b, c, raw_rmse, log_rmse):
    equation = format_fit_equation(a, b, c)
    return (
        f"<b>{label}</b><br>"
        f"{equation}<br>"
        f"Raw RMSE: {raw_rmse:.2f}<br>"
        f"Log-RMSE: {log_rmse:.3f}<br>"
        "Params: %{x:.2f}B<br>"
        "Predicted CR: %{y:.2f}%<extra></extra>"
    )


def build_fit_summary_legend_text(label, a, b, c, raw_rmse, log_rmse, include_label=True):
    parts = []
    if include_label and label:
        parts.append(label)
    parts.append(format_fit_equation(a, b, c))
    parts.append(f"Raw RMSE: {raw_rmse:.2f}")
    parts.append(f"Log-RMSE: {log_rmse:.3f}")
    return "<br>".join(parts)


def read_about_md():
    with open("about.md", "r", encoding="utf-8") as f:
        return f.read()


def read_longctx_about_md():
    with open("longctx_about.md", "r", encoding="utf-8") as f:
        return f.read()


def update_table(
    data_manager: DataManager,
    period: str,
    models_size: list,
    metric: str,
    visible_columns: list,
    color_columns: list,
    size_range: list,
    midpoint: float = 0.5,
    ascending: bool = True,
    request: gr.Request = None,
):
    is_dark_mode = request.is_dark if request else False
    print(
        f"Updating - time: {datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')}, period: {period}, models: {models_size}, metric: {metric}, visible_columns: {visible_columns}, color_columns: {color_columns}, size_range: {size_range}, ascending: {ascending}, is_dark: {is_dark_mode}\n"
    )

    target_file_name = [model_size_to_file_name[model] for model in models_size]
    metric_code = metric_to_sheet[metric]

    # 过滤掉不在当前 period 可用列中的列名，避免错误
    if visible_columns:
        available_columns = data_manager.get_available_columns(period)
        visible_columns = [col for col in visible_columns if col in available_columns]

    filtered_data = data_manager.query(
        period=period,
        metric_code=metric_code,
        param_range=(size_range[0], size_range[1]),
        model_groups=target_file_name,
        visible_columns=visible_columns,
    )

    if len(filtered_data) == 0:
        return "No data available for the selected models and period."

    colors = ["#2ca02c", "#2b2b2b", "#d62728"] if is_dark_mode else ["#63be7b", "#ffffff", "#f8696b"]

    vmin, vmax, vmid = {}, {}, {}
    for column in filtered_data.columns:
        if column in ["Name", "Params (B)"]:
            continue
        col_values = filtered_data[column].dropna()
        if len(col_values) > 1:
            sorted_values = np.sort(col_values)
            vmin[column] = sorted_values.min()
            vmax[column] = sorted_values.max()
            idx = int(len(sorted_values) * midpoint)
            vmid[column] = sorted_values[idx]

    def custom_background_gradient(series, cmap, vmin_val, vmax_val, vmid_val):
        if len(series) == 0:
            return series

        def normalize(x):
            if pd.isna(x):
                return 0.5  # Neutral for NaN
            if vmid_val == vmin_val and x <= vmid_val:
                return 0.0
            if vmid_val == vmax_val and x >= vmid_val:
                return 1.0
            if vmid_val == vmin_val or vmid_val == vmax_val:
                return 0.5
            if x <= vmid_val:
                return 0.5 * (x - vmin_val) / (vmid_val - vmin_val)
            else:
                return 0.5 + 0.5 * (x - vmid_val) / (vmax_val - vmid_val)

        normed = series.apply(normalize)
        cmap_colors = [cmap(x) for x in normed]
        return ["background-color: rgba({}, {}, {}, {}); color: black;".format(*[int(255 * c) for c in color[:3]], color[3]) for color in cmap_colors]

    target_color_columns = []
    if "Average" in color_columns:
        target_color_columns.append("Average (lower=better)")
    if "Individual Tests" in color_columns:
        target_color_columns.extend([col for col in filtered_data.columns if col not in ["Name", "Params (B)", "Average (lower=better)"]])

    def color_params_column_dynamic(value):
        if not pd.notna(value):
            return "default"

        if is_dark_mode:
            return "background-color: #4b4936; color: #f0f0f0;"
        else:
            return "background-color: #fffdd0; color: black;"

    tooltip_map_by_column = {}
    for column in filtered_data.columns:
        if column in ["Name", "Params (B)"]:
            continue

        valid_mask = filtered_data[column].notna()
        valid_count = int(valid_mask.sum())
        if valid_count == 0:
            continue

        ranks = filtered_data[column].rank(method="min", ascending=ascending, na_option="bottom").astype("Int64")
        tooltip_map_by_column[column] = {
            value: f"Rank: {int(rank)}/{valid_count}" for value, rank in zip(filtered_data.loc[valid_mask, column], ranks.loc[valid_mask])
        }

    def format_cell_content(text, tooltip=None, extra_classes=None):
        safe_text = html.escape(str(text))
        classes = ["cell-tooltip-trigger"]
        if extra_classes:
            classes.extend(extra_classes)

        tooltip_attr = ""
        if tooltip:
            tooltip_attr = f' data-tooltip="{html.escape(str(tooltip))}"'

        return f'<span class="{" ".join(classes)}"{tooltip_attr}>{safe_text}</span>'

    def truncate_model_name(text):
        if len(text) <= MODEL_NAME_DISPLAY_MAX_CHARS:
            return text

        cutoff = text.rfind("-", 0, MODEL_NAME_DISPLAY_MAX_CHARS + 1)
        if cutoff > 0:
            return text[:cutoff]
        return text[:MODEL_NAME_DISPLAY_MAX_CHARS]

    def format_model_name(value):
        if not pd.notna(value):
            return ""

        full_value = str(value)
        safe_value = html.escape(full_value)
        safe_display_value = html.escape(truncate_model_name(full_value))
        return (
            f'<span class="cell-tooltip-trigger model-name-trigger" data-tooltip="{safe_value}">'
            f'<span class="model-name-cell">{safe_display_value}</span>'
            "</span>"
        )

    formatter = {}
    for column in filtered_data.columns:
        if column == "Name":
            formatter[column] = format_model_name
        elif filtered_data[column].dtype in ["float64", "float32"]:
            if column == "Params (B)":
                formatter[column] = lambda value: "" if not pd.notna(value) else f"{value:.3f}"
            else:
                tooltip_map = tooltip_map_by_column.get(column, {})

                def make_numeric_formatter(current_tooltip_map):
                    def format_numeric(value):
                        if not pd.notna(value):
                            return ""

                        return format_cell_content(f"{value:.3f}", tooltip=current_tooltip_map.get(value))

                    return format_numeric

                formatter[column] = make_numeric_formatter(tooltip_map)
    styler = filtered_data.style.format(formatter)
    styler = styler.map(color_params_column_dynamic, subset=["Params (B)"])
    for column in target_color_columns:
        if column in vmin:
            custom_cmap = LinearSegmentedColormap.from_list("custom_cmap", colors)
            styler = styler.apply(
                custom_background_gradient, cmap=custom_cmap, vmin_val=vmin[column], vmax_val=vmax[column], vmid_val=vmid[column], subset=[column]
            )

    styler = styler.hide(axis="index")
    widths = [250, 85, 85, 70, 70, 70, 70, 70, 70, 70, 70, 70, 70, 70, 70, 70, 70, 70]
    hover_filter = "brightness(1.06) saturate(1.03)" if is_dark_mode else "brightness(0.97) saturate(1.05)"
    hover_shadow = "inset 0 0 0 2px rgba(255, 255, 255, 0.45)" if is_dark_mode else "inset 0 0 0 2px rgba(0, 0, 0, 0.16)"

    table_styles = []
    table_styles.append(
        {
            "selector": "th",
            "props": [
                ("background-color", "var(--background-fill-secondary)"),
                ("color", "var(--body-text-color)"),
                ("padding", "6px 4px"),
                ("font-weight", "bold"),
                ("font-size", "12px"),
                ("line-height", "1.2"),
                ("white-space", "normal"),
                ("word-break", "break-word"),
                ("overflow-wrap", "anywhere"),
            ],
        }
    )
    table_styles.append(
        {
            "selector": "td",
            "props": [
                ("position", "relative"),
                ("transition", "filter 0.12s ease, box-shadow 0.12s ease"),
            ],
        }
    )
    table_styles.append(
        {
            "selector": "td:hover",
            "props": [
                ("filter", hover_filter),
                ("box-shadow", hover_shadow),
                ("z-index", "1"),
            ],
        }
    )
    for i, w in enumerate(widths):
        table_styles.append(
            {
                "selector": f"th.col{i}, td.col{i}",
                "props": [
                    ("min-width", f"{w}px"),
                    ("max-width", f"{w}px"),
                    ("text-align", "center"),
                    ("border", f"1px solid var(--border-color-primary)"),
                ],
            }
        )
    table_styles.append(
        {
            "selector": "td.col0 .model-name-cell",
            "props": [
                ("display", "block"),
                ("width", "100%"),
                ("overflow", "hidden"),
                ("text-overflow", "ellipsis"),
                ("white-space", "nowrap"),
            ],
        }
    )
    table_styles.append(
        {
            "selector": "td .cell-tooltip-trigger",
            "props": [
                ("display", "block"),
                ("position", "relative"),
                ("width", "100%"),
            ],
        }
    )
    table_styles.append(
        {
            "selector": "td:hover .cell-tooltip-trigger[data-tooltip]::after",
            "props": [
                ("content", "attr(data-tooltip)"),
                ("position", "absolute"),
                ("top", "calc(100% + 6px)"),
                ("left", "50%"),
                ("transform", "translateX(-50%)"),
                ("z-index", "10000"),
                ("background-color", "var(--background-fill-secondary)"),
                ("color", "var(--body-text-color)"),
                ("border", "1px solid var(--border-color-primary)"),
                ("border-radius", "6px"),
                ("padding", "4px 6px"),
                ("font-size", "12px"),
                ("white-space", "nowrap"),
                ("pointer-events", "none"),
                ("box-shadow", "0 4px 14px rgba(0, 0, 0, 0.14)"),
            ],
        }
    )
    table_styles.append(
        {
            "selector": "td:last-child:hover .cell-tooltip-trigger[data-tooltip]::after",
            "props": [
                ("left", "auto"),
                ("right", "0"),
                ("transform", "none"),
            ],
        }
    )
    table_styles.append(
        {
            "selector": "tbody tr:last-child td:hover .cell-tooltip-trigger[data-tooltip]::after",
            "props": [
                ("top", "auto"),
                ("bottom", "calc(100% + 6px)"),
            ],
        }
    )
    styler = styler.set_table_styles(table_styles)
    styler = styler.set_table_attributes(
        'style="border-collapse: collapse; border: 1px solid var(--border-color-primary); width: max-content !important; min-width: 100% !important; margin-left: 0 !important; margin-right: 0 !important;"'
    )
    table_html = styler.to_html()
    return (
        '<div class="leaderboard-table-scroll" style="width: 100%; max-width: 100%; overflow-x: auto; padding-bottom: 4px;">'
        f"{table_html}"
        "</div>"
    )


def check_model_exists(model_id):
    api = HfApi()
    try:
        model_info = api.model_info(model_id)
        return "Exists and is accessible"
    except GatedRepoError:
        return "Exists but is restricted"
    except HfHubHTTPError as e:
        if e.response.status_code == 404:
            return "Does not exist"
        else:
            return "Error: " + str(e)


def submit_model(name):
    if "Exists" not in check_model_exists(name):
        return f"# ERROR: Model {name} does not exist on Hugging Face!"
    try:
        response = requests.post(webhook_url, json={"content": name})
        if response.status_code == 200:
            response_data = response.json()
            if response_data.get("status") == "success":
                return "# SUCCESS: We will check the model as soon as possible. Thank you for your submission!"
            else:
                return f"# ERROR: {response_data.get('message', 'Unknown error')}"
        else:
            return f"# ERROR: Failed to submit model {name}. Server returned status code {response.status_code}."
    except requests.exceptions.HTTPError:
        return "# ERROR: Network error while contacting queue. Please try again in a few minutes."
    except Exception as e:
        print(e)
        return "ERROR: Unexpected error. Please try again later."


def power_law_with_offset(x, a, b, c):
    """带偏置的幂律函数: y = a * x^b + c"""
    return a * np.power(x, b) + c


def calculate_fit_delta_percent(x_values, y_values, params):
    """Return percent delta from fit; negative is better because lower CR is better."""
    x_arr = np.array(x_values, dtype=float)
    y_arr = np.array(y_values, dtype=float)
    fit_y = power_law_with_offset(x_arr, *params)
    with np.errstate(divide="ignore", invalid="ignore"):
        delta = ((y_arr - fit_y) / fit_y) * 100
    delta[~np.isfinite(delta)] = np.nan
    return delta.tolist()


def create_fit_delta_colorscale(cmin, cmax):
    if cmax <= 0:
        return [[0.0, FIT_RESIDUAL_NEUTRAL_COLOR], [1.0, FIT_RESIDUAL_GOOD_COLOR]]
    if cmin >= 0:
        return [[0.0, FIT_RESIDUAL_BAD_COLOR], [1.0, FIT_RESIDUAL_NEUTRAL_COLOR]]

    mapped_cmin = -cmax
    mapped_cmax = -cmin
    zero_position = (0 - mapped_cmin) / (mapped_cmax - mapped_cmin)
    return [
        [0.0, FIT_RESIDUAL_BAD_COLOR],
        [zero_position, FIT_RESIDUAL_NEUTRAL_COLOR],
        [1.0, FIT_RESIDUAL_GOOD_COLOR],
    ]


def create_fit_delta_color_values(delta_values):
    return [-float(v) if np.isfinite(v) else np.nan for v in delta_values]


def get_fit_delta_bounds(delta_values):
    finite_values = [float(v) for v in delta_values if np.isfinite(v)]
    if not finite_values:
        return -1.0, 1.0

    cmin = min(finite_values)
    cmax = max(finite_values)

    if cmin == cmax:
        if cmin < 0:
            cmax = 0.0
        elif cmin > 0:
            cmin = 0.0
        else:
            cmin, cmax = -1.0, 1.0
    else:
        cmin = min(cmin, 0.0)
        cmax = max(cmax, 0.0)
    return cmin, cmax


def _hex_to_rgb(hex_color):
    hex_color = hex_color.lstrip("#")
    return tuple(int(hex_color[i : i + 2], 16) for i in (0, 2, 4))


def _interpolate_rgb(start_rgb, end_rgb, t):
    t = min(max(float(t), 0.0), 1.0)
    return tuple(int(round(start + (end - start) * t)) for start, end in zip(start_rgb, end_rgb))


def fit_delta_to_rgba(delta_value, cmin, cmax, alpha=0.22):
    if not np.isfinite(delta_value):
        return "transparent"

    good_rgb = _hex_to_rgb(FIT_RESIDUAL_GOOD_COLOR)
    neutral_rgb = _hex_to_rgb(FIT_RESIDUAL_NEUTRAL_COLOR)
    bad_rgb = _hex_to_rgb(FIT_RESIDUAL_BAD_COLOR)

    if cmax <= 0:
        denom = cmax - cmin
        t = 1.0 if denom == 0 else (cmax - float(delta_value)) / denom
        rgb = _interpolate_rgb(neutral_rgb, good_rgb, t)
    elif cmin >= 0:
        denom = cmax - cmin
        t = 1.0 if denom == 0 else (float(delta_value) - cmin) / denom
        rgb = _interpolate_rgb(neutral_rgb, bad_rgb, t)
    elif float(delta_value) <= 0:
        denom = 0.0 - cmin
        t = 1.0 if denom == 0 else abs(float(delta_value)) / denom
        rgb = _interpolate_rgb(neutral_rgb, good_rgb, t)
    else:
        denom = cmax - 0.0
        t = 1.0 if denom == 0 else float(delta_value) / denom
        rgb = _interpolate_rgb(neutral_rgb, bad_rgb, t)

    return f"rgba({rgb[0]}, {rgb[1]}, {rgb[2]}, {alpha:.3f})"


def create_fit_delta_coloraxis(delta_values):
    cmin, cmax = get_fit_delta_bounds(delta_values)
    tick_values = np.linspace(cmax, cmin, 5)
    return dict(
        colorscale=create_fit_delta_colorscale(cmin, cmax),
        cmin=-cmax,
        cmax=-cmin,
        showscale=False,
        colorbar=dict(
            title="vs fit",
            tickmode="array",
            tickvals=[-float(v) for v in tick_values],
            ticktext=[f"{float(v):+.1f}%" for v in tick_values],
        ),
    )


def filter_pareto_frontier(x_values, y_values, names):
    """
    筛选帕累托前沿的点
    对于每一个数据点 (x_i, y_i)，如果不存在另一个点 (x_j, y_j) 满足 x_j <= x_i 且 y_j < y_i，
    那么这个点 (x_i, y_i) 就属于帕累托前沿。

    参数:
        x_values: 参数量列表
        y_values: 压缩比列表
        names: 模型名称列表

    返回:
        (pareto_x, pareto_y, pareto_names): 帕累托前沿的点
    """
    points = list(zip(x_values, y_values, names))
    pareto_points = []

    for i, (xi, yi, ni) in enumerate(points):
        is_pareto = True
        for j, (xj, yj, _) in enumerate(points):
            if i != j:
                # 如果存在另一个点，参数量更小或相等，且压缩比更低，则当前点不在帕累托前沿
                if xj <= xi and yj < yi:
                    is_pareto = False
                    break
        if is_pareto:
            pareto_points.append((xi, yi, ni))

    if pareto_points:
        pareto_x, pareto_y, pareto_names = zip(*pareto_points)
        return list(pareto_x), list(pareto_y), list(pareto_names)
    else:
        return [], [], []


def _build_frontier_table_rows(x_values, y_values, names, use_pareto=False):
    valid_data = []
    for x, y, n in zip(x_values, y_values, names):
        try:
            x_float = float(x)
            y_float = float(y)
        except (TypeError, ValueError):
            continue
        if x_float > 0 and y_float > 0 and not np.isnan(x_float) and not np.isnan(y_float):
            valid_data.append((x_float, y_float, str(n)))

    if not valid_data:
        return [], []

    valid_x, valid_y, valid_names = zip(*valid_data)
    valid_x, valid_y, valid_names = list(valid_x), list(valid_y), list(valid_names)
    pareto_x, pareto_y, pareto_names = filter_pareto_frontier(valid_x, valid_y, valid_names)

    fit_x, fit_y, fit_names = (pareto_x, pareto_y, pareto_names) if use_pareto else (valid_x, valid_y, valid_names)
    if len(fit_x) < 2:
        return [], []

    params, _, _, _, _ = fit_power_law_with_offset(
        fit_x,
        fit_y,
        extrapolate_max_b=SCALING_EXTRAPOLATE_MAX_B,
    )
    point_delta_values = calculate_fit_delta_percent(valid_x, valid_y, params)
    pareto_delta_values = calculate_fit_delta_percent(pareto_x, pareto_y, params)

    rows = [
        {
            "params": round(float(param), 3),
            "model": _format_frontier_model_name(model),
            "ratio%": round(float(ratio), 3),
            "vs fit%": float(delta),
        }
        for param, ratio, model, delta in zip(pareto_x, pareto_y, pareto_names, pareto_delta_values)
    ]
    return sorted(rows, key=lambda row: (-row["params"], row["model"])), point_delta_values


def _format_frontier_model_name(model_name: str):
    if not isinstance(model_name, str):
        return str(model_name)

    lower_name = model_name.lower()
    ctx_suffix = "-ctx8192"
    if lower_name.startswith("rwkv") and lower_name.endswith(ctx_suffix):
        return model_name[: -len(ctx_suffix)]
    return model_name


def create_scaling_frontier_table(
    data_manager: DataManager,
    period: str,
    mode: str,
    selected_datasets: list,
    display_mode: str,
    use_pareto: bool = False,
):
    new_df = data_manager.query(
        period=period,
        metric_code="cr",
        param_range=(0, 40),
        model_groups=None,
        visible_columns=None,
    )

    if len(new_df) == 0:
        return pd.DataFrame(columns=FRONTIER_TABLE_COLUMNS)

    rows = []
    all_delta_values = []
    is_by_dataset = "By Dataset" in str(mode)

    if not is_by_dataset:
        series_rows, series_deltas = _build_frontier_table_rows(
            new_df["Params (B)"].astype(float).tolist(),
            new_df["Average (lower=better)"].astype(float).tolist(),
            new_df["Name"].tolist(),
            use_pareto=use_pareto,
        )
        rows.extend(series_rows)
        all_delta_values.extend(series_deltas)
        return _finalize_frontier_table(rows, all_delta_values)

    selected_datasets = selected_datasets or []
    selected_datasets = [dataset for dataset in selected_datasets if dataset in new_df.columns]
    if not selected_datasets:
        return pd.DataFrame(columns=FRONTIER_TABLE_COLUMNS)

    x_values = new_df["Params (B)"].astype(float).tolist()
    names = new_df["Name"].tolist()

    if display_mode == "average":
        avg_y_values = []
        for i in range(len(new_df)):
            valid_values = []
            for dataset in selected_datasets:
                val = new_df[dataset].iloc[i]
                if pd.notna(val) and val > 0:
                    valid_values.append(val)
            avg_y_values.append(float(np.mean(valid_values)) if valid_values else np.nan)

        series_rows, series_deltas = _build_frontier_table_rows(x_values, avg_y_values, names, use_pareto=use_pareto)
        rows.extend(series_rows)
        all_delta_values.extend(series_deltas)
    else:
        for dataset in selected_datasets:
            series_rows, series_deltas = _build_frontier_table_rows(
                x_values,
                new_df[dataset].astype(float).tolist(),
                names,
                use_pareto=use_pareto,
            )
            rows.extend(series_rows)
            all_delta_values.extend(series_deltas)

    best_rows = {}
    for row in rows:
        row_key = (row["params"], row["model"])
        if row_key not in best_rows or row["ratio%"] < best_rows[row_key]["ratio%"]:
            best_rows[row_key] = row
    rows = sorted(best_rows.values(), key=lambda row: (-row["params"], row["model"]))
    return _finalize_frontier_table(rows, all_delta_values)


def _finalize_frontier_table(rows, all_delta_values):
    if not rows:
        return pd.DataFrame(columns=FRONTIER_TABLE_COLUMNS)

    cmin, cmax = get_fit_delta_bounds(all_delta_values)
    for row in rows:
        delta = row.get("vs fit%", np.nan)
        row["row_bg"] = fit_delta_to_rgba(delta, cmin, cmax, alpha=0.22)
        row["row_hover_bg"] = fit_delta_to_rgba(delta, cmin, cmax, alpha=0.32)
    return pd.DataFrame(rows, columns=FRONTIER_TABLE_COLUMNS)


def render_scaling_frontier_table(frontier_df):
    if frontier_df is None or len(frontier_df) == 0:
        rows_html = '<tr><td class="empty" colspan="4">No Pareto frontier models</td></tr>'
    else:
        rows = []
        for _, row in frontier_df.iterrows():
            params = html.escape(f'{float(row["params"]):.3f}')
            model = html.escape(str(row["model"]))
            ratio = html.escape(f'{float(row["ratio%"]):.3f}')
            fit_delta = row.get("vs fit%", np.nan)
            fit_delta_text = html.escape(f"{float(fit_delta):+.2f}%") if np.isfinite(fit_delta) else "--"
            row_bg = html.escape(str(row.get("row_bg", "transparent")), quote=True)
            row_hover_bg = html.escape(str(row.get("row_hover_bg", row_bg)), quote=True)
            rows.append(
                f'<tr style="--row-bg: {row_bg}; --row-hover-bg: {row_hover_bg};">'
                f'<td class="params">{params}</td>'
                f'<td class="model">{model}</td>'
                f'<td class="ratio">{ratio}</td>'
                f'<td class="delta">{fit_delta_text}</td>'
                f"</tr>"
            )
        rows_html = "\n".join(rows)

    return f"""
<div class="frontier-table-card">
  <div class="frontier-table-title">Pareto Frontier Models</div>
  <div class="frontier-table-scroll">
    <table class="frontier-table-inner">
      <colgroup>
        <col class="params-col">
        <col class="model-col">
        <col class="ratio-col">
        <col class="delta-col">
      </colgroup>
      <thead>
        <tr><th>params</th><th>model</th><th>ratio%</th><th>vs fit</th></tr>
      </thead>
      <tbody>
        {rows_html}
      </tbody>
    </table>
  </div>
</div>
"""


def fit_power_law_with_offset(x_values, y_values, extrapolate_max_b=None, num_points=SCALING_FIT_POINTS):
    """
    使用带偏置的幂律拟合原始数据
    返回: (params, raw_rmse, log_rmse, fit_x, fit_y)
    """
    x_arr = np.array(x_values)
    y_arr = np.array(y_values)

    # 初始参数估计
    # 使用简单的幂律拟合作为初始值
    log_x = np.log10(x_arr)
    log_y = np.log10(y_arr)
    slope, intercept = np.polyfit(log_x, log_y, 1)

    a_init = 10**intercept
    b_init = slope
    c_init = 0  # 偏置初始值设为0
    x_min, x_max = x_arr.min(), x_arr.max()
    x_start = max(x_min * 0.8, np.finfo(float).tiny)
    x_end = x_max * 1.2
    if extrapolate_max_b is not None:
        x_end = max(x_end, extrapolate_max_b)
    fit_x = np.logspace(np.log10(x_start), np.log10(x_end), num_points)

    try:
        # 使用curve_fit进行非线性拟合
        params, _ = curve_fit(power_law_with_offset, x_arr, y_arr, p0=[a_init, b_init, c_init], maxfev=10000)
        a, b, c = params

        # 计算预测值
        y_pred = power_law_with_offset(x_arr, a, b, c)

        # 计算原始空间 RMSE
        raw_rmse = np.sqrt(np.mean((y_arr - y_pred) ** 2))

        # 计算对数空间 RMSE
        log_y_actual = np.log10(y_arr)
        log_y_pred = np.log10(y_pred)
        log_rmse = np.sqrt(np.mean((log_y_actual - log_y_pred) ** 2))

        # 生成拟合曲线的点
        fit_y = power_law_with_offset(fit_x, a, b, c)

        return params, raw_rmse, log_rmse, fit_x, fit_y
    except Exception as e:
        print(f"Fitting failed: {e}")
        # 如果拟合失败，返回简单幂律拟合结果
        a = a_init
        b = b_init
        c = 0
        params = (a, b, c)

        y_pred = a * np.power(x_arr, b)

        # 计算原始空间 RMSE
        raw_rmse = np.sqrt(np.mean((y_arr - y_pred) ** 2))

        # 计算对数空间 RMSE
        log_y_actual = np.log10(y_arr)
        log_y_pred = np.log10(y_pred)
        log_rmse = np.sqrt(np.mean((log_y_actual - log_y_pred) ** 2))

        fit_y = a * np.power(fit_x, b)

        return params, raw_rmse, log_rmse, fit_x, fit_y


def create_scaling_plot(data_manager: DataManager, period: str, use_pareto: bool = False):
    new_df = data_manager.query(
        period=period,
        metric_code="cr",
        param_range=(0, 40),
        model_groups=None,
        visible_columns=None,
    )

    if len(new_df) == 0:
        fig = go.Figure()
        fig.update_layout(
            title={"text": "Compression Ratio Scaling Law", "x": 0.5},
            width=SCALING_PLOT_WIDTH,
            height=SCALING_PLOT_HEIGHT,
            margin=SCALING_PLOT_MARGIN,
        )
        return fig

    x_values = new_df["Params (B)"].astype(float).tolist()
    y_values = new_df["Average (lower=better)"].astype(float).tolist()
    names = new_df["Name"].tolist()

    # 过滤掉无效值（NaN, 0, 负数）
    valid_data = [(x, y, n) for x, y, n in zip(x_values, y_values, names) if x > 0 and y > 0 and not np.isnan(x) and not np.isnan(y)]
    if len(valid_data) == 0:
        fig = go.Figure()
        fig.update_layout(
            title={"text": "Compression Ratio Scaling Law", "x": 0.5},
            width=SCALING_PLOT_WIDTH,
            height=SCALING_PLOT_HEIGHT,
            margin=SCALING_PLOT_MARGIN,
        )
        return fig

    x_values, y_values, names = zip(*valid_data)
    x_values, y_values, names = list(x_values), list(y_values), list(names)

    # 如果选择帕累托前沿，筛选数据点
    if use_pareto:
        fit_x_values, fit_y_values, fit_names = filter_pareto_frontier(x_values, y_values, names)
        if len(fit_x_values) == 0:
            fig = go.Figure()
            fig.update_layout(
                title={"text": "Compression Ratio Scaling Law - No Pareto Frontier", "x": 0.5},
                width=SCALING_PLOT_WIDTH,
                height=SCALING_PLOT_HEIGHT,
                margin=SCALING_PLOT_MARGIN,
            )
            return fig
    else:
        fit_x_values, fit_y_values, fit_names = x_values, y_values, names

    x_min_val = min(x_values)
    x_max_val = max(x_values)
    x_axis_max = x_max_val

    # 使用筛选后的数据进行拟合
    params, raw_rmse, log_rmse, fit_x, fit_y = fit_power_law_with_offset(
        fit_x_values,
        fit_y_values,
        extrapolate_max_b=SCALING_EXTRAPOLATE_MAX_B,
    )
    a, b, c = params
    y_min_val = min(y_values)
    y_max_val = max(y_values)
    positive_fit_y = fit_y[fit_y > 0]
    if positive_fit_y.size:
        y_min_val = min(y_min_val, float(positive_fit_y.min()))
        y_max_val = max(y_max_val, float(positive_fit_y.max()))

    x_min = np.log10(x_min_val)
    x_max = np.log10(x_axis_max)
    y_min = np.log10(y_min_val)
    y_max = np.log10(y_max_val)
    x_dtick = (x_max - x_min) / 4
    y_dtick = (y_max - y_min) / 4

    fig = go.Figure()
    point_delta_values = calculate_fit_delta_percent(x_values, y_values, params)

    # Pareto 模式下，普通散点仅显示非 Pareto 点，避免同一点重复绘制
    base_points = list(zip(x_values, y_values, names, point_delta_values))
    if use_pareto:
        pareto_points = set(zip(fit_x_values, fit_y_values, fit_names))
        base_points = [p for p in base_points if (p[0], p[1], p[2]) not in pareto_points]

    if base_points:
        base_x_values, base_y_values, base_names, base_delta_values = zip(*base_points)
        fig.add_trace(
            go.Scatter(
                x=list(base_x_values),
                y=list(base_y_values),
                mode="markers",
                name="Non-Pareto Models" if use_pareto else "All Models",
                marker=dict(size=12, color=create_fit_delta_color_values(base_delta_values), coloraxis="coloraxis", opacity=0.85),
                text=list(base_names),
                customdata=list(zip(base_x_values, base_y_values, base_delta_values)),
                hovertemplate=(
                    "<b>%{text}</b><br>"
                    + "Params: %{customdata[0]:.2f}B<br>"
                    + "Compression Ratio: %{customdata[1]:.2f}%<br>"
                    + "vs Fit: %{customdata[2]:+.2f}%<br>"
                    + "<extra></extra>"
                ),
            )
        )

    # 如果使用帕累托前沿，高亮显示帕累托前沿的点
    if use_pareto:
        pareto_delta_values = calculate_fit_delta_percent(fit_x_values, fit_y_values, params)
        fig.add_trace(
            go.Scatter(
                x=fit_x_values,
                y=fit_y_values,
                mode="markers",
                name="Pareto Frontier",
                marker=dict(
                    size=14,
                    color=create_fit_delta_color_values(pareto_delta_values),
                    coloraxis="coloraxis",
                    symbol="diamond",
                    opacity=1.0,
                    line=dict(color="#263238", width=1),
                ),
                text=fit_names,
                customdata=list(zip(fit_x_values, fit_y_values, pareto_delta_values)),
                hovertemplate=(
                    "<b>%{text}</b> (Pareto)<br>"
                    + "Params: %{customdata[0]:.2f}B<br>"
                    + "Compression Ratio: %{customdata[1]:.2f}%<br>"
                    + "vs Fit: %{customdata[2]:+.2f}%<br>"
                    + "<extra></extra>"
                ),
            )
        )

    # 添加拟合曲线
    fit_type = "Pareto Fit" if use_pareto else "Fit"
    fig.add_trace(
        go.Scatter(
            x=fit_x.tolist(),
            y=fit_y.tolist(),
            mode="lines",
            name=build_fit_summary_legend_text(fit_type, a, b, c, raw_rmse, log_rmse),
            line=dict(color=FIT_LINE_COLOR, width=2, dash="dash"),
            hovertemplate=build_fit_line_hovertemplate(fit_type, a, b, c, raw_rmse, log_rmse),
        )
    )

    title_suffix = " (Pareto Frontier)" if use_pareto else ""
    fig.update_layout(
        title={"text": f"Compression Ratio Scaling Law{title_suffix}", "x": 0.5, "xanchor": "center", "yanchor": "top"},
        width=SCALING_PLOT_WIDTH,
        height=SCALING_PLOT_HEIGHT,
        margin=SCALING_PLOT_MARGIN,
        showlegend=True,
        legend=dict(
            yanchor="top",
            y=0.99,
            xanchor="left",
            x=0.01,
            bgcolor="rgba(255,255,255,0.8)",
        ),
        xaxis=dict(
            title="Parameters (B)",
            showgrid=True,
            zeroline=False,
            type="log",
            dtick=x_dtick,
            tickformat=".2f",
            range=[x_min - 0.1, x_max + 0.1],
        ),
        yaxis=dict(
            title="Compression Ratio (%)",
            showgrid=True,
            zeroline=False,
            type="log",
            dtick=y_dtick,
            tickformat=".2f",
            range=[y_min - 0.1, y_max + 0.1],
            autorange="reversed",
        ),
        coloraxis=create_fit_delta_coloraxis(point_delta_values),
    )
    return fig


def create_category_scaling_plot(
    data_manager: DataManager, period: str, selected_datasets: list, display_mode: str = "separate", use_pareto: bool = False
):
    """
    为选中的数据集绘制 scaling law 拟合线
    display_mode: "separate" - 每个数据集单独显示, "average" - 计算选中数据集的平均值
    use_pareto: True - 只拟合帕累托前沿的点, False - 拟合所有数据点
    """
    new_df = data_manager.query(
        period=period,
        metric_code="cr",
        param_range=(0, 40),
        model_groups=None,
        visible_columns=None,
    )

    if len(new_df) == 0 or not selected_datasets:
        fig = go.Figure()
        fig.update_layout(
            title={"text": "Scaling Law by Dataset", "x": 0.5},
            width=SCALING_PLOT_WIDTH,
            height=700,
            margin=SCALING_PLOT_MARGIN,
        )
        return fig

    # 颜色配色方案 - 使用高对比度、饱和度高的颜色
    color_palette = [
        "#1f77b4",  # 蓝色
        "#ff7f0e",  # 橙色
        "#2ca02c",  # 绿色
        "#d62728",  # 红色
        "#9467bd",  # 紫色
        "#8c564b",  # 棕色
        "#e377c2",  # 粉色
        "#17becf",  # 青色
        "#bcbd22",  # 黄绿色
        "#7f7f7f",  # 灰色
    ]

    fig = go.Figure()

    # 用于计算全局坐标范围
    all_x_values = []
    all_y_values = []
    all_delta_values = []
    if display_mode == "average":
        # 平均模式：计算选中数据集的平均值
        x_values = new_df["Params (B)"].astype(float).tolist()
        names = new_df["Name"].tolist()

        # 计算每个模型在选中数据集上的平均值
        avg_y_values = []
        for i in range(len(new_df)):
            valid_values = []
            for dataset in selected_datasets:
                if dataset in new_df.columns:
                    val = new_df[dataset].iloc[i]
                    if not np.isnan(val) and val > 0:
                        valid_values.append(val)
            if valid_values:
                avg_y_values.append(np.mean(valid_values))
            else:
                avg_y_values.append(np.nan)

        # 过滤掉无效值
        valid_data = [(x, y, n) for x, y, n in zip(x_values, avg_y_values, names) if x > 0 and not np.isnan(y) and y > 0]

        if len(valid_data) >= 2:
            x_vals, y_vals, name_vals = zip(*valid_data)
            x_vals, y_vals, name_vals = list(x_vals), list(y_vals), list(name_vals)

            all_x_values.extend(x_vals)
            all_y_values.extend(y_vals)

            color = "#39C5BB"

            # 如果选择帕累托前沿，筛选数据点
            if use_pareto:
                fit_x_vals, fit_y_vals, fit_name_vals = filter_pareto_frontier(x_vals, y_vals, name_vals)
                if len(fit_x_vals) < 2:
                    return fig
            else:
                fit_x_vals, fit_y_vals, fit_name_vals = x_vals, y_vals, name_vals

            # 使用筛选后的数据进行拟合
            params, raw_rmse, log_rmse, fit_x, fit_y = fit_power_law_with_offset(
                fit_x_vals,
                fit_y_vals,
                extrapolate_max_b=SCALING_EXTRAPOLATE_MAX_B,
            )
            a, b, c = params
            positive_fit_y = fit_y[fit_y > 0]
            if positive_fit_y.size:
                all_y_values.extend(positive_fit_y.tolist())
            point_delta_values = calculate_fit_delta_percent(x_vals, y_vals, params)
            all_delta_values.extend(point_delta_values)

            # 构建数据集名称列表（用于hover显示）
            datasets_label = f"Average of {len(selected_datasets)} datasets"

            # Pareto 模式下，普通散点仅显示非 Pareto 点，避免同一点重复绘制
            base_points = list(zip(x_vals, y_vals, name_vals, point_delta_values))
            if use_pareto:
                pareto_points = set(zip(fit_x_vals, fit_y_vals, fit_name_vals))
                base_points = [p for p in base_points if (p[0], p[1], p[2]) not in pareto_points]

            if base_points:
                base_x_vals, base_y_vals, base_name_vals, base_delta_vals = zip(*base_points)
                fig.add_trace(
                    go.Scatter(
                        x=list(base_x_vals),
                        y=list(base_y_vals),
                        mode="markers",
                        name=f"{datasets_label} (Non-Pareto)" if use_pareto else datasets_label,
                        marker=dict(size=12, color=create_fit_delta_color_values(base_delta_vals), coloraxis="coloraxis", opacity=0.85),
                        text=list(base_name_vals),
                        customdata=list(zip(base_x_vals, base_y_vals, base_delta_vals)),
                        hovertemplate=(
                            f"<b>%{{text}}</b><br>{datasets_label}<br>"
                            + "Params: %{customdata[0]:.2f}B<br>"
                            + "CR: %{customdata[1]:.2f}%<br>"
                            + "vs Fit: %{customdata[2]:+.2f}%<br>"
                            + "<extra></extra>"
                        ),
                    )
                )

            # 如果使用帕累托前沿，高亮显示帕累托前沿的点
            if use_pareto:
                pareto_delta_vals = calculate_fit_delta_percent(fit_x_vals, fit_y_vals, params)
                fig.add_trace(
                    go.Scatter(
                        x=fit_x_vals,
                        y=fit_y_vals,
                        mode="markers",
                        name="Pareto Frontier",
                        marker=dict(
                            size=14,
                            color=create_fit_delta_color_values(pareto_delta_vals),
                            coloraxis="coloraxis",
                            symbol="diamond",
                            opacity=1.0,
                            line=dict(color="#263238", width=1),
                        ),
                        text=fit_name_vals,
                        customdata=list(zip(fit_x_vals, fit_y_vals, pareto_delta_vals)),
                        hovertemplate=(
                            f"<b>%{{text}}</b> (Pareto)<br>{datasets_label}<br>"
                            + "Params: %{customdata[0]:.2f}B<br>"
                            + "CR: %{customdata[1]:.2f}%<br>"
                            + "vs Fit: %{customdata[2]:+.2f}%<br>"
                            + "<extra></extra>"
                        ),
                    )
                )

            # 添加拟合曲线
            fit_type = "Pareto Fit" if use_pareto else "Fit"
            fit_label = f"{datasets_label} ({fit_type})"
            fit_legend_name = fit_label
            if len(selected_datasets) == 1:
                fit_legend_name = build_fit_summary_legend_text(fit_label, a, b, c, raw_rmse, log_rmse)
            fig.add_trace(
                go.Scatter(
                    x=fit_x.tolist(),
                    y=fit_y.tolist(),
                    mode="lines",
                    name=fit_legend_name,
                    line=dict(color=FIT_LINE_COLOR, width=2, dash="dash"),
                    hovertemplate=build_fit_line_hovertemplate(fit_label, a, b, c, raw_rmse, log_rmse),
                )
            )
    else:
        # 单独显示模式：为每个数据集创建散点图和拟合线
        fig.add_trace(
            go.Scatter(
                x=[None],
                y=[None],
                mode="markers",
                name="Non-Pareto",
                marker=dict(size=9, color="rgba(55,65,81,0.9)", symbol="circle"),
                hoverinfo="skip",
            )
        )
        if use_pareto:
            fig.add_trace(
                go.Scatter(
                    x=[None],
                    y=[None],
                    mode="markers",
                    name="Pareto",
                    marker=dict(
                        size=10,
                        color="rgba(255,255,255,0.95)",
                        symbol="diamond",
                        line=dict(color="#263238", width=1.1),
                    ),
                    hoverinfo="skip",
                )
            )

        for idx, dataset in enumerate(selected_datasets):
            if dataset not in new_df.columns:
                continue

            # 提取该数据集的数据
            x_values = new_df["Params (B)"].astype(float).tolist()
            y_values = new_df[dataset].astype(float).tolist()
            names = new_df["Name"].tolist()

            # 过滤掉无效值
            valid_data = [(x, y, n) for x, y, n in zip(x_values, y_values, names) if x > 0 and y > 0 and not np.isnan(x) and not np.isnan(y)]

            if len(valid_data) < 2:  # 至少需要2个点才能拟合
                continue

            x_vals, y_vals, name_vals = zip(*valid_data)
            x_vals, y_vals, name_vals = list(x_vals), list(y_vals), list(name_vals)

            all_x_values.extend(x_vals)
            all_y_values.extend(y_vals)

            color = color_palette[idx % len(color_palette)]

            # 如果选择帕累托前沿，筛选数据点
            if use_pareto:
                fit_x_vals, fit_y_vals, fit_name_vals = filter_pareto_frontier(x_vals, y_vals, name_vals)
                if len(fit_x_vals) < 2:
                    continue
            else:
                fit_x_vals, fit_y_vals, fit_name_vals = x_vals, y_vals, name_vals

            # 使用筛选后的数据进行拟合
            params, raw_rmse, log_rmse, fit_x, fit_y = fit_power_law_with_offset(
                fit_x_vals,
                fit_y_vals,
                extrapolate_max_b=SCALING_EXTRAPOLATE_MAX_B,
            )
            a, b, c = params
            positive_fit_y = fit_y[fit_y > 0]
            if positive_fit_y.size:
                all_y_values.extend(positive_fit_y.tolist())
            point_delta_values = calculate_fit_delta_percent(x_vals, y_vals, params)
            all_delta_values.extend(point_delta_values)

            # Pareto 模式下，普通散点仅显示非 Pareto 点，避免同一点重复绘制
            base_points = list(zip(x_vals, y_vals, name_vals, point_delta_values))
            if use_pareto:
                pareto_points = set(zip(fit_x_vals, fit_y_vals, fit_name_vals))
                base_points = [p for p in base_points if (p[0], p[1], p[2]) not in pareto_points]

            if base_points:
                base_x_vals, base_y_vals, base_name_vals, base_delta_vals = zip(*base_points)
                fig.add_trace(
                    go.Scatter(
                        x=list(base_x_vals),
                        y=list(base_y_vals),
                        mode="markers",
                        name=f"{dataset} (Non-Pareto)" if use_pareto else f"{dataset}",
                        marker=dict(size=10, color=create_fit_delta_color_values(base_delta_vals), coloraxis="coloraxis", opacity=0.8),
                        text=list(base_name_vals),
                        customdata=list(zip(base_x_vals, base_y_vals, base_delta_vals)),
                        hovertemplate=(
                            f"<b>%{{text}}</b><br>{dataset}<br>"
                            + "Params: %{customdata[0]:.2f}B<br>"
                            + "CR: %{customdata[1]:.2f}%<br>"
                            + "vs Fit: %{customdata[2]:+.2f}%<br>"
                            + "<extra></extra>"
                        ),
                        legendgroup=dataset,
                        showlegend=False,
                    )
                )

            # 如果使用帕累托前沿，高亮显示帕累托前沿的点
            if use_pareto:
                pareto_delta_vals = calculate_fit_delta_percent(fit_x_vals, fit_y_vals, params)
                fig.add_trace(
                    go.Scatter(
                        x=fit_x_vals,
                        y=fit_y_vals,
                        mode="markers",
                        name=f"{dataset} (Pareto)",
                        marker=dict(
                            size=12,
                            color=create_fit_delta_color_values(pareto_delta_vals),
                            coloraxis="coloraxis",
                            symbol="diamond",
                            opacity=1.0,
                            line=dict(color="#263238", width=1),
                        ),
                        text=fit_name_vals,
                        customdata=list(zip(fit_x_vals, fit_y_vals, pareto_delta_vals)),
                        hovertemplate=(
                            f"<b>%{{text}}</b> (Pareto)<br>{dataset}<br>"
                            + "Params: %{customdata[0]:.2f}B<br>"
                            + "CR: %{customdata[1]:.2f}%<br>"
                            + "vs Fit: %{customdata[2]:+.2f}%<br>"
                            + "<extra></extra>"
                        ),
                        legendgroup=dataset,
                        showlegend=False,
                    )
                )

            # 添加拟合曲线
            fit_type = "Pareto Fit" if use_pareto else "Fit"
            fit_label = f"{dataset} ({fit_type})"
            fit_legend_name = dataset
            if len(selected_datasets) == 1:
                fit_legend_name = build_fit_summary_legend_text(dataset, a, b, c, raw_rmse, log_rmse)
            fig.add_trace(
                go.Scatter(
                    x=fit_x.tolist(),
                    y=fit_y.tolist(),
                    mode="lines",
                    name=fit_legend_name,
                    line=dict(color=color, width=2, dash="dash"),
                    hovertemplate=build_fit_line_hovertemplate(fit_label, a, b, c, raw_rmse, log_rmse),
                    legendgroup=dataset,
                    showlegend=True,
                )
            )

    if not all_x_values or not all_y_values:
        fig = go.Figure()
        fig.update_layout(
            title={"text": "Scaling Law by Dataset - No Valid Data", "x": 0.5},
            width=SCALING_PLOT_WIDTH,
            height=700,
            margin=SCALING_PLOT_MARGIN,
        )
        return fig

    # 计算全局坐标范围
    x_min_val = min(all_x_values)
    x_max_val = max(all_x_values)
    x_axis_max = x_max_val
    x_min, x_max = np.log10(x_min_val), np.log10(x_axis_max)
    y_min, y_max = np.log10(min(all_y_values)), np.log10(max(all_y_values))
    x_dtick = (x_max - x_min) / 4
    y_dtick = (y_max - y_min) / 4
    coloraxis = create_fit_delta_coloraxis(all_delta_values)
    coloraxis["colorbar"].update(
        dict(
            x=1.02,
            xanchor="left",
            y=0.5,
            yanchor="middle",
            len=0.72,
            thickness=28,
            tickfont=dict(size=9),
            title=dict(text="vs fit", side="top"),
        )
    )

    fig.update_layout(
        title={"text": "Scaling Law by Dataset", "x": 0.5, "xanchor": "center", "yanchor": "top"},
        width=SCALING_PLOT_WIDTH,
        height=700,
        showlegend=True,
        legend=dict(
            yanchor="top",
            y=0.98,
            xanchor="left",
            x=0.02,
            bgcolor="rgba(255,255,255,0.78)",
            bordercolor="rgba(15,23,42,0.08)",
            borderwidth=1,
            font=dict(size=9),
            tracegroupgap=2,
        ),
        xaxis=dict(
            title="Parameters (B)",
            showgrid=True,
            zeroline=False,
            type="log",
            dtick=x_dtick,
            tickformat=".2f",
            range=[x_min - 0.1, x_max + 0.1],
        ),
        yaxis=dict(
            title="Compression Ratio (%)",
            showgrid=True,
            zeroline=False,
            type="log",
            dtick=y_dtick,
            tickformat=".2f",
            range=[y_min - 0.1, y_max + 0.1],
            autorange="reversed",
        ),
        coloraxis=coloraxis,
        margin=dict(l=70, r=80, t=70, b=65),
    )
    return fig


if __name__ == "__main__":
    data_manager = DataManager("data")
    time_list = data_manager.get_available_periods()
    last_period = time_list[0]

    # Long Context Data
    lc_dm = LongContextDataManager("longctx_data")
    lc_periods = lc_dm.get_available_periods()
    default_lc_period = lc_periods[0]
    MODE_ABS_AVG = "Absolute (Averaged by Model)"
    MODE_ABS_SINGLE = "Absolute (By Dataset)"
    MODE_REL_AVG = "Relative (Averaged by Model)"
    MODE_REL_SINGLE = "Relative (By Dataset)"
    lc_modes = [MODE_ABS_AVG, MODE_ABS_SINGLE, MODE_REL_AVG, MODE_REL_SINGLE]
    default_lc_mode = MODE_ABS_AVG
    # init_lc_choices = lc_dm.get_model_choices(default_lc_period)
    init_lc_choices = lc_dm.get_model_choices(default_lc_period)

    def get_default_model(choices):
        """获取默认模型，优先选择 Qwen3-8B-Base，否则返回第一个模型"""
        if not choices:
            return None
        for display_name, model_name in choices:
            if model_name == "Qwen3-8B-Base":
                return model_name
        return choices[0][1]

    def create_initial_lc_plot():
        if not init_lc_choices:
            return None
        default_model = get_default_model(init_lc_choices)
        data_map = {}
        paths = lc_dm.get_paths_for_model(default_lc_period, default_model)
        data_map[default_model] = paths
        return draw_long_context_plot(default_lc_mode, data_map, None, 0.1, 32, 32, [None, None], 0.02)

    initial_lc_plot = create_initial_lc_plot()

    initial_fig = create_scaling_plot(data_manager, last_period) if last_period else go.Figure()
    initial_metric = metric_list[0]
    initial_columns = data_manager.get_available_columns(last_period)
    initial_colors = ["Average", "Individual Tests"]
    initial_size_range = [0, 50]
    # 默认不显示 ao3 nonenglish 列
    default_visible_columns = [c for c in initial_columns if c != "ao3 nonenglish"]
    initial_data = update_table(
        data_manager, last_period, model_size_list, initial_metric, default_visible_columns, initial_colors, initial_size_range
    )

    theme = gr.themes.Default(
        font=[
            gr.themes.GoogleFont("Source Sans Pro"),
            "ui-sans-serif",
            "system-ui",
            "sans-serif",
        ],
        font_mono=[
            "IBM Plex Mono",
            "ui-monospace",
            "Consolas",
            "monospace",
        ],
    )
    with gr.Blocks(theme=theme, css=css) as demo:
        gr.HTML(TITLE_HTML)
        gr.HTML(SUBTITLE_HTML)
        gr.HTML(LINKS_HTML)
        with gr.Tabs() as tabs:
            with gr.Tab("🏆 Leaderboard"):
                with gr.Row():
                    with gr.Column():
                        period_selector = gr.Dropdown(label="Period", choices=time_list, value=last_period)
                        metric_selector = gr.Dropdown(label="Metric", choices=metric_list, value=initial_metric)
                        model_selector = gr.CheckboxGroup(label="Model Size", choices=model_size_list, value=model_size_list)
                        size_range_slider = RangeSlider(minimum=0, maximum=50, value=[0, 50], step=0.1, label="Model Size Range")
                        midpoint_slider = gr.Slider(minimum=0.1, maximum=0.9, value=0.5, step=0.01, label="Color Gradient Midpoint")
                        color_selector = gr.CheckboxGroup(label="Colored Columns", choices=["Average", "Individual Tests"], value=initial_colors)
                    with gr.Column():
                        # Data Source 分组定义
                        code_cols = ["github cpp", "github javascript", "github python", "github markdown", "github other"]
                        science_cols = ["arxiv math", "arxiv physics", "arxiv cs", "arxiv other", "biorxiv all"]
                        knowledge_cols = ["wikipedia english", "bbc news", "ao3 english"]
                        multilingual_cols = ["wikipedia nonenglish", "ao3 nonenglish"]

                        initial_code = [c for c in code_cols if c in initial_columns]
                        initial_science = [c for c in science_cols if c in initial_columns]
                        initial_knowledge = [c for c in knowledge_cols if c in initial_columns]
                        initial_multilingual = [c for c in multilingual_cols if c in initial_columns]
                        default_multilingual = [c for c in initial_multilingual if c != "ao3 nonenglish"]

                        with gr.Column(elem_classes=["data-source-box"]):
                            gr.Markdown("Data Sources")

                            # 代码 (Code)
                            with gr.Row():
                                toggle_code = gr.Checkbox(label="💻 Code", value=True, scale=0, min_width=150)
                                colfilter_code = gr.CheckboxGroup(
                                    choices=initial_code, value=initial_code, show_label=False, scale=3, elem_classes=["aligned-checkboxes"]
                                )

                            # 科学 (Science)
                            with gr.Row():
                                toggle_science = gr.Checkbox(label="🔬 Science", value=True, scale=0, min_width=150)
                                colfilter_science = gr.CheckboxGroup(
                                    choices=initial_science, value=initial_science, show_label=False, scale=3, elem_classes=["aligned-checkboxes"]
                                )

                            # 世界知识 (Knowledge)
                            with gr.Row():
                                toggle_knowledge = gr.Checkbox(label="📖 Knowledge", value=True, scale=0, min_width=150)
                                colfilter_knowledge = gr.CheckboxGroup(
                                    choices=initial_knowledge, value=initial_knowledge, show_label=False, scale=3, elem_classes=["aligned-checkboxes"]
                                )

                            # 多语言 (Multilingual)
                            with gr.Row():
                                toggle_multilingual = gr.Checkbox(label="🌍 Multilingual", value=True, scale=0, min_width=150)
                                colfilter_multilingual = gr.CheckboxGroup(
                                    choices=initial_multilingual,
                                    value=default_multilingual,
                                    show_label=False,
                                    scale=3,
                                    elem_classes=["aligned-checkboxes"],
                                )

                table = gr.HTML(initial_data, elem_classes=["leaderboard-table"])

                def update_table_wrapper(
                    period, models_size, metric, code_sel, science_sel, knowledge_sel, multilingual_sel, color_columns, size_range, midpoint
                ):
                    visible_columns = code_sel + science_sel + knowledge_sel + multilingual_sel
                    return update_table(data_manager, period, models_size, metric, visible_columns, color_columns, size_range, midpoint)

                def update_column_choices(period, cur_code, cur_science, cur_knowledge, cur_multilingual):
                    if not period:
                        empty = gr.update(choices=[], value=[])
                        return empty, empty, empty, empty
                    columns = data_manager.get_available_columns(period)

                    new_code = [c for c in code_cols if c in columns]
                    new_science = [c for c in science_cols if c in columns]
                    new_knowledge = [c for c in knowledge_cols if c in columns]
                    new_multilingual = [c for c in multilingual_cols if c in columns]

                    sel_code = [c for c in cur_code if c in new_code] if cur_code else new_code
                    sel_science = [c for c in cur_science if c in new_science] if cur_science else new_science
                    sel_knowledge = [c for c in cur_knowledge if c in new_knowledge] if cur_knowledge else new_knowledge
                    sel_multilingual = [c for c in cur_multilingual if c in new_multilingual] if cur_multilingual else new_multilingual

                    if not sel_code:
                        sel_code = new_code
                    if not sel_science:
                        sel_science = new_science
                    if not sel_knowledge:
                        sel_knowledge = new_knowledge
                    if not sel_multilingual:
                        sel_multilingual = new_multilingual

                    return (
                        gr.update(choices=new_code, value=sel_code),
                        gr.update(choices=new_science, value=sel_science),
                        gr.update(choices=new_knowledge, value=sel_knowledge),
                        gr.update(choices=new_multilingual, value=sel_multilingual),
                    )

                # 总开关功能
                def toggle_group(enabled, group_cols, available_cols):
                    valid_cols = [c for c in group_cols if c in available_cols]
                    return valid_cols if enabled else []

                toggle_code.change(lambda enabled: toggle_group(enabled, code_cols, initial_columns), inputs=[toggle_code], outputs=[colfilter_code])
                toggle_science.change(
                    lambda enabled: toggle_group(enabled, science_cols, initial_columns), inputs=[toggle_science], outputs=[colfilter_science]
                )
                toggle_knowledge.change(
                    lambda enabled: toggle_group(enabled, knowledge_cols, initial_columns), inputs=[toggle_knowledge], outputs=[colfilter_knowledge]
                )
                toggle_multilingual.change(
                    lambda enabled: toggle_group(enabled, multilingual_cols, initial_columns),
                    inputs=[toggle_multilingual],
                    outputs=[colfilter_multilingual],
                )

                shared_inputs = [
                    period_selector,
                    model_selector,
                    metric_selector,
                    colfilter_code,
                    colfilter_science,
                    colfilter_knowledge,
                    colfilter_multilingual,
                    color_selector,
                    size_range_slider,
                    midpoint_slider,
                ]

                period_selector.change(
                    update_column_choices,
                    inputs=[period_selector, colfilter_code, colfilter_science, colfilter_knowledge, colfilter_multilingual],
                    outputs=[colfilter_code, colfilter_science, colfilter_knowledge, colfilter_multilingual],
                )
                period_selector.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                model_selector.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                metric_selector.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                colfilter_code.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                colfilter_science.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                colfilter_knowledge.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                colfilter_multilingual.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                color_selector.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                size_range_slider.change(update_table_wrapper, inputs=shared_inputs, outputs=table)
                midpoint_slider.change(update_table_wrapper, inputs=shared_inputs, outputs=table)

            with gr.Tab("📚 Long Context"):
                gr.Markdown(read_longctx_about_md())
                with gr.Row():
                    with gr.Column(scale=1):
                        lc_period_dropdown = gr.Dropdown(label="Period", choices=lc_periods, value=default_lc_period)

                        lc_mode_radio = gr.Radio(label="Visualization Mode", choices=lc_modes, value=default_lc_mode)

                        gr.Markdown("### Model / Dataset Selection")

                        default_model = get_default_model(init_lc_choices)
                        default_selected_models = [default_model] if default_model else []
                        lc_select_abs = gr.Dropdown(
                            label="Select Models", choices=init_lc_choices, value=default_selected_models, multiselect=True, visible=True
                        )

                        lc_select_base = gr.Dropdown(
                            label="Baseline Model",
                            choices=init_lc_choices,
                            value=None,
                            multiselect=False,
                            visible=False,
                        )

                        lc_select_comp = gr.Dropdown(label="Comparison Models", choices=init_lc_choices, value=[], multiselect=True, visible=False)

                        # By Dataset mode selectors
                        init_dataset_choices = lc_dm.get_dataset_choices(default_lc_period) if default_lc_period else []
                        default_selected_datasets = [init_dataset_choices[0][1]] if init_dataset_choices else []
                        lc_select_datasets = gr.Dropdown(
                            label="Select Datasets", choices=init_dataset_choices, value=default_selected_datasets, multiselect=True, visible=False
                        )

                        lc_select_models_single = gr.Dropdown(
                            label="Select Models", choices=init_lc_choices, value=default_selected_models, multiselect=True, visible=False
                        )

                        lc_select_base_model_single = gr.Dropdown(
                            label="Baseline Model",
                            choices=init_lc_choices,
                            value=None,
                            multiselect=False,
                            visible=False,
                        )

                        lc_select_comp_models_single = gr.Dropdown(
                            label="Comparison Models", choices=init_lc_choices, value=[], multiselect=True, visible=False
                        )

                        with gr.Accordion("Advanced Settings", open=True):
                            lc_smooth = gr.Slider(1, 125, 32, step=1, label="Smooth Window")
                            lc_cutoff = gr.Slider(0.05, 1.0, 0.1, step=0.05, label="Cutoff Ratio")
                            lc_tail_drop = gr.Slider(0.0, 0.9, 0.02, step=0.01, label="Tail Drop Ratio")
                            lc_offset = gr.Number(32, label="Start Offset (Bytes)")
                            with gr.Row():
                                lc_ymin = gr.Textbox(label="Y Min", placeholder="Auto", value="")
                                lc_ymax = gr.Textbox(label="Y Max", placeholder="Auto", value="")

                        lc_btn_plot = gr.Button("Visualize", variant="primary")

                    with gr.Column(scale=3):
                        lc_plot_output = gr.Plot(label="Visualization Result", value=initial_lc_plot)

                def update_lc_inputs(period, mode):
                    if not period:
                        return tuple([gr.update()] * 7)

                    is_model_agg = "Averaged by Model" in mode
                    is_single_dataset = "By Dataset" in mode
                    is_relative = "Relative" in mode

                    def get_default_model(choices):
                        """获取默认模型，优先选择 Qwen3-8B-Base，否则返回第一个模型"""
                        if not choices:
                            return None
                        for display_name, model_name in choices:
                            if model_name == "Qwen3-8B-Base":
                                return model_name
                        return choices[0][1] if choices else None

                    if is_model_agg:
                        # Averaged by Model mode - use existing logic
                        choices = lc_dm.get_model_choices(period)
                        label_suffix = "Models"
                        if not is_relative:
                            # Absolute (Averaged by Model) - 默认选择 Qwen3-8B-Base
                            default_model = get_default_model(choices)
                            default_selected = [default_model] if default_model else []
                            return (
                                gr.update(visible=True, choices=choices, label=f"Select {label_suffix}", value=default_selected),
                                gr.update(visible=False, choices=choices, value=None),
                                gr.update(visible=False, choices=choices, value=[]),
                                gr.update(visible=False, value=[]),
                                gr.update(visible=False, value=[]),
                                gr.update(visible=False, value=None),
                                gr.update(visible=False, value=[]),
                            )
                        else:
                            default_baseline = get_default_model(choices)
                            return (
                                gr.update(visible=False, choices=choices, value=[]),
                                gr.update(visible=True, choices=choices, label=f"Baseline", value=default_baseline),
                                gr.update(visible=True, choices=choices, label=f"Comparison", value=[]),
                                gr.update(visible=False, value=[]),
                                gr.update(visible=False, value=[]),
                                gr.update(visible=False, value=None),
                                gr.update(visible=False, value=[]),
                            )
                    else:
                        # By Dataset mode
                        dataset_choices = lc_dm.get_dataset_choices(period)
                        model_choices = lc_dm.get_model_choices(period)
                        if not is_relative:
                            # Absolute By Dataset - 默认选择 Qwen3-8B-Base
                            default_model = get_default_model(model_choices)
                            default_selected = [default_model] if default_model else []
                            return (
                                gr.update(visible=False, value=[]),
                                gr.update(visible=False, value=None),
                                gr.update(visible=False, value=[]),
                                gr.update(visible=True, choices=dataset_choices, value=[]),
                                gr.update(visible=True, choices=model_choices, value=default_selected),
                                gr.update(visible=False, value=None),
                                gr.update(visible=False, value=[]),
                            )
                        else:
                            # Relative By Dataset - use same datasets for all models
                            default_baseline = get_default_model(model_choices)
                            return (
                                gr.update(visible=False, value=[]),
                                gr.update(visible=False, value=None),
                                gr.update(visible=False, value=[]),
                                gr.update(visible=True, choices=dataset_choices, value=[]),
                                gr.update(visible=False, value=[]),
                                gr.update(visible=True, choices=model_choices, value=default_baseline),
                                gr.update(visible=True, choices=model_choices, value=[]),
                            )

                lc_period_dropdown.change(
                    fn=update_lc_inputs,
                    inputs=[lc_period_dropdown, lc_mode_radio],
                    outputs=[
                        lc_select_abs,
                        lc_select_base,
                        lc_select_comp,
                        lc_select_datasets,
                        lc_select_models_single,
                        lc_select_base_model_single,
                        lc_select_comp_models_single,
                    ],
                )
                lc_mode_radio.change(
                    fn=update_lc_inputs,
                    inputs=[lc_period_dropdown, lc_mode_radio],
                    outputs=[
                        lc_select_abs,
                        lc_select_base,
                        lc_select_comp,
                        lc_select_datasets,
                        lc_select_models_single,
                        lc_select_base_model_single,
                        lc_select_comp_models_single,
                    ],
                )

                def run_lc_plot(
                    mode,
                    period,
                    sel_abs,
                    sel_base,
                    sel_comp,
                    sel_datasets,
                    sel_models_single,
                    sel_base_model_single,
                    sel_comp_models_single,
                    smooth,
                    cutoff,
                    tail_drop,
                    offset,
                    ymin,
                    ymax,
                ):
                    data_map = {}
                    baseline_key = None

                    is_model_agg = "Averaged by Model" in mode
                    is_relative = "Relative" in mode

                    if is_model_agg:
                        # Averaged by Model mode - existing logic
                        if not is_relative:
                            selection = sel_abs
                        else:
                            if not sel_base:
                                return None
                            selection = [sel_base] + sel_comp
                            baseline_key = sel_base

                        if not selection:
                            return None

                        for item in selection:
                            paths = lc_dm.get_paths_for_model(period, item)
                            if paths:
                                data_map[item] = paths
                    else:
                        # By Dataset mode
                        if not is_relative:
                            # Absolute By Dataset
                            if not sel_datasets or not sel_models_single:
                                return None

                            for model_name in sel_models_single:
                                paths = lc_dm.get_paths_for_model_and_datasets(period, model_name, sel_datasets)
                                if paths:
                                    data_map[model_name] = paths
                        else:
                            # Relative By Dataset - use same datasets for all models
                            if not sel_datasets or not sel_base_model_single:
                                return None

                            # Baseline model with selected datasets (averaged)
                            baseline_paths = lc_dm.get_paths_for_model_and_datasets(period, sel_base_model_single, sel_datasets)
                            if baseline_paths:
                                baseline_key = sel_base_model_single
                                data_map[baseline_key] = baseline_paths

                            # Comparison models with same datasets (averaged)
                            if sel_comp_models_single:
                                for model_name in sel_comp_models_single:
                                    paths = lc_dm.get_paths_for_model_and_datasets(period, model_name, sel_datasets)
                                    if paths:
                                        data_map[model_name] = paths

                    if not data_map:
                        return None

                    def _to_float_or_none(val):
                        if val is None:
                            return None
                        s = str(val).strip()
                        if not s:
                            return None
                        try:
                            return float(s)
                        except ValueError:
                            return None

                    ymin = _to_float_or_none(ymin)
                    ymax = _to_float_or_none(ymax)

                    y_range = [ymin, ymax]

                    return draw_long_context_plot(mode, data_map, baseline_key, cutoff, smooth, int(offset), y_range, tail_drop)

                lc_btn_plot.click(
                    fn=run_lc_plot,
                    inputs=[
                        lc_mode_radio,
                        lc_period_dropdown,
                        lc_select_abs,
                        lc_select_base,
                        lc_select_comp,
                        lc_select_datasets,
                        lc_select_models_single,
                        lc_select_base_model_single,
                        lc_select_comp_models_single,
                        lc_smooth,
                        lc_cutoff,
                        lc_tail_drop,
                        lc_offset,
                        lc_ymin,
                        lc_ymax,
                    ],
                    outputs=lc_plot_output,
                )

            with gr.Tab("📈 Scaling Law"):
                gr.Markdown("### Compression Ratio Scaling Law")
                gr.Markdown("Explore how compression ratio scales with model parameters across different datasets.")

                # 显示模式选择
                MODE_OVERALL = "📊 Overall (Average)"
                MODE_BY_DATASET = "📈 By Dataset"
                scaling_modes = [MODE_OVERALL, MODE_BY_DATASET]

                # 数据集列表
                all_datasets = [
                    "github cpp",
                    "github javascript",
                    "github python",
                    "github markdown",
                    "github other",
                    "arxiv math",
                    "arxiv physics",
                    "arxiv cs",
                    "arxiv other",
                    "biorxiv all",
                    "wikipedia english",
                    "wikipedia nonenglish",
                    "bbc news",
                    "ao3 english",
                    "ao3 nonenglish",
                ]
                initial_datasets = all_datasets[:4]

                with gr.Row():
                    with gr.Column(scale=1):
                        scaling_period_selector = gr.Dropdown(label="Period", choices=time_list, value=last_period)
                        scaling_mode_radio = gr.Radio(label="Display Mode", choices=scaling_modes, value=MODE_OVERALL)
                        # 拟合方式选择
                        scaling_fit_mode = gr.Radio(
                            label="Fitting Method",
                            choices=[("Fit All Data", False), ("Fit Pareto Frontier", True)],
                            value=True,
                            info="Pareto Frontier: only fit points where no other model has both smaller parameters and lower compression ratio",
                        )
                        # 数据集选择器（初始隐藏）
                        scaling_dataset_selector = gr.CheckboxGroup(
                            label="Select Datasets", choices=all_datasets, value=initial_datasets, visible=False
                        )
                        # 数据集显示方式（初始隐藏）
                        scaling_dataset_display_mode = gr.Radio(
                            label="Dataset Display",
                            choices=[("Average Selected", "average"), ("Show Separately", "separate")],
                            value="separate",
                            visible=False,
                        )

                    with gr.Column(scale=5, min_width=1440):
                        with gr.Row(equal_height=False):
                            with gr.Column(scale=4, min_width=1040):
                                initial_scaling_fig = create_scaling_plot(data_manager, last_period, use_pareto=True) if last_period else go.Figure()
                                scaling_plot = gr.Plot(initial_scaling_fig, elem_classes=["scaling-plot"])
                            with gr.Column(scale=1, min_width=380):
                                initial_scaling_frontier_df = (
                                    create_scaling_frontier_table(data_manager, last_period, MODE_OVERALL, initial_datasets, "separate", use_pareto=True)
                                    if last_period
                                    else pd.DataFrame(columns=FRONTIER_TABLE_COLUMNS)
                                )
                                scaling_frontier_table = gr.HTML(
                                    value=render_scaling_frontier_table(initial_scaling_frontier_df),
                                    elem_classes=["frontier-table"],
                                )

                def update_scaling_mode_visibility(mode):
                    """根据模式切换数据集选择器的可见性"""
                    is_by_dataset = mode == MODE_BY_DATASET
                    return gr.update(visible=is_by_dataset), gr.update(visible=is_by_dataset)

                def update_scaling_plot_unified(period, mode, datasets, dataset_display_mode, use_pareto):
                    """统一的绑图更新函数"""
                    if mode == MODE_OVERALL:
                        fig = create_scaling_plot(data_manager, period, use_pareto)
                    else:  # MODE_BY_DATASET
                        fig = create_category_scaling_plot(data_manager, period, datasets, dataset_display_mode, use_pareto)

                    frontier_table = create_scaling_frontier_table(
                        data_manager,
                        period,
                        mode,
                        datasets,
                        dataset_display_mode,
                        use_pareto=use_pareto,
                    )
                    return fig, gr.update(value=render_scaling_frontier_table(frontier_table), visible=bool(use_pareto))

                # 模式切换时更新可见性和图表
                scaling_mode_radio.change(
                    fn=update_scaling_mode_visibility, inputs=[scaling_mode_radio], outputs=[scaling_dataset_selector, scaling_dataset_display_mode]
                )
                scaling_mode_radio.change(
                    fn=update_scaling_plot_unified,
                    inputs=[scaling_period_selector, scaling_mode_radio, scaling_dataset_selector, scaling_dataset_display_mode, scaling_fit_mode],
                    outputs=[scaling_plot, scaling_frontier_table],
                )

                # Period 改变时更新图表
                scaling_period_selector.change(
                    fn=update_scaling_plot_unified,
                    inputs=[scaling_period_selector, scaling_mode_radio, scaling_dataset_selector, scaling_dataset_display_mode, scaling_fit_mode],
                    outputs=[scaling_plot, scaling_frontier_table],
                )

                # 数据集选择改变时更新图表
                scaling_dataset_selector.change(
                    fn=update_scaling_plot_unified,
                    inputs=[scaling_period_selector, scaling_mode_radio, scaling_dataset_selector, scaling_dataset_display_mode, scaling_fit_mode],
                    outputs=[scaling_plot, scaling_frontier_table],
                )

                # 数据集显示模式改变时更新图表
                scaling_dataset_display_mode.change(
                    fn=update_scaling_plot_unified,
                    inputs=[scaling_period_selector, scaling_mode_radio, scaling_dataset_selector, scaling_dataset_display_mode, scaling_fit_mode],
                    outputs=[scaling_plot, scaling_frontier_table],
                )

                # 拟合方式改变时更新图表
                scaling_fit_mode.change(
                    fn=update_scaling_plot_unified,
                    inputs=[scaling_period_selector, scaling_mode_radio, scaling_dataset_selector, scaling_dataset_display_mode, scaling_fit_mode],
                    outputs=[scaling_plot, scaling_frontier_table],
                )

            with gr.Tab("ℹ️ About"):
                gr.Markdown(read_about_md())

            with gr.Tab("🚀 Submit"):
                with gr.Group():
                    with gr.Row():
                        model_name = gr.Textbox(max_lines=1, placeholder="Enter model name...", show_label=False, scale=4)
                        submit = gr.Button("Submit", variant="primary", scale=0)
                output = gr.Markdown("# Enter a public HF repo id, then hit Submit to add it to the evaluation queue.")
                submit.click(fn=submit_model, inputs=model_name, outputs=output)

    demo.launch(share=False)