Spaces:
Running
Running
| """Reusable UI components for the MSEB Gradio Leaderboard.""" | |
| from __future__ import annotations | |
| import gradio as gr | |
| import pandas as pd | |
| from src.config import TaskSpec | |
| def create_header() -> gr.HTML: | |
| """Renders the top banner, title, description, and benchmark statistics.""" | |
| header_html = """ | |
| <div class="header-container"> | |
| <h1 class="header-title">MSEB Leaderboard</h1> | |
| <p class="header-subtitle"> | |
| <strong>Massive Sound Embedding Benchmark (MSEB)</strong> — | |
| A comprehensive multilingual benchmark for evaluating speech, audio, | |
| and multimodal models. See <a href="https://github.com/google-research/mseb" target="_blank">git</a>/<a href="https://neurips.cc/virtual/2025/loc/san-diego/poster/121597" target="_blank">poster</a>/<a href="https://proceedings.neurips.cc/paper_files/paper/2025/file/2c878aec1c052a835511c262033ad348-Paper-Datasets_and_Benchmarks_Track.pdf" target="_blank">paper</a> for more details. | |
| </p> | |
| </div> | |
| """ | |
| return gr.HTML(header_html) | |
| def create_search_bar( | |
| placeholder: str = "🔍 Search models by name (e.g., 'gemini', 'flash', 'asr=truth')...", | |
| label: str = "Search Models", | |
| ) -> gr.Textbox: | |
| """Creates a text input component for dynamic model name filtering.""" | |
| return gr.Textbox( | |
| show_label=False, | |
| placeholder=placeholder, | |
| label=label, | |
| elem_classes=["search-input"], | |
| interactive=True, | |
| ) | |
| def create_column_selector( | |
| choices: list[str], | |
| default_selected: list[str], | |
| label: str = "Select Columns to Display:", | |
| show_label: bool = True, | |
| ) -> gr.CheckboxGroup: | |
| """Creates an interactive checkbox group for toggling visible columns.""" | |
| return gr.CheckboxGroup( | |
| choices=choices, | |
| value=default_selected, | |
| label=label, | |
| show_label=show_label, | |
| elem_classes=["column-selector-group"], | |
| interactive=True, | |
| ) | |
| def create_leaderboard_table( | |
| df: pd.DataFrame, | |
| ) -> gr.DataFrame: | |
| """Creates a Gradio DataFrame configured for tabular leaderboard presentation. | |
| Supports clickable Markdown model links and formatted numeric columns. | |
| Args: | |
| df: DataFrame containing leaderboard data. | |
| Returns: | |
| gr.DataFrame configured for display. | |
| """ | |
| return gr.DataFrame( | |
| value=df, | |
| interactive=False, | |
| datatype="markdown", | |
| elem_classes=["leaderboard-table"], | |
| max_height=768, | |
| wrap=True, | |
| ) | |
| def create_task_header( | |
| task_spec: TaskSpec, | |
| dataset_count: int = 0, | |
| ) -> gr.HTML: | |
| """Renders task description and primary metric info callout banner.""" | |
| symbol = task_spec.metric_direction.symbol | |
| direction_desc = ( | |
| "Higher is better" | |
| if task_spec.metric_direction.name == "HIGHER_IS_BETTER" | |
| else "Lower is better" | |
| ) | |
| unit_desc = ( | |
| "Percentage [0–100%]" | |
| if task_spec.default_unit == "percentage" | |
| else "Raw distance metric" | |
| ) | |
| html = f""" | |
| <div class="task-info-banner"> | |
| <div class="task-info-title">{task_spec.display_name} ({symbol} {direction_desc})</div> | |
| <div class="task-info-desc"> | |
| {task_spec.description} • | |
| <strong>Primary Metric:</strong> <code>{task_spec.primary_metric}</code> ({unit_desc}) • | |
| <strong>Evaluated Datasets:</strong> {dataset_count} | |
| </div> | |
| </div> | |
| """ | |
| return gr.HTML(html) | |
| def create_inactive_task_notice(task_spec: TaskSpec) -> gr.HTML: | |
| """Renders a notice for tasks with no evaluation records in the active results catalog.""" | |
| html = f""" | |
| <div class="inactive-task-banner"> | |
| ⚠️ <strong>Note:</strong> No evaluated models are currently available in | |
| <code>results/</code> for <strong>{task_spec.display_name}</strong>. | |
| Showing the canonical column schema. | |
| </div> | |
| """ | |
| return gr.HTML(html) | |