app.py

import gradio as gr
import pandas as pd
import json

from src.about import (
    REPRODUCIBILITY_TEXT,
    INTRODUCTION_TEXT,
    ABOUT_TEXT,
    TITLE,
)
from src.display.css_html_js import custom_css, custom_js
from src.display.formatting import make_clickable_field

def build_leaderboard(type):
    with open('data/results.json', 'r') as f:
        results = json.load(f)

    with open('data/tasks.json', 'r') as f:
        tasks = json.load(f)

    # Filter tasks based on type
    filtered_tasks = {k: v for k, v in tasks.items() if v['type'] == type}

    data = []
    for model_name, model_data in results.items():
        # For agentic type, skip models that have all null values for agentic tasks
        if type == "agentic":
            has_agentic_results = any(
                model_data['results'].get(task, {}).get(tasks[task]['metric']) is not None 
                for task in filtered_tasks
            )
            if not has_agentic_results:
                continue

        model_sha = model_data["config"]["model_sha"]
        model_name = model_data["config"]["model_name"]
        row = {
            'Model': make_clickable_field(model_name, model_sha)
        }
        
        for dataset, metrics in model_data['results'].items():
            # Only include metrics for tasks of the specified type
            if dataset in filtered_tasks:
                value = next(iter(metrics.values()))
                log_url = metrics.get('log_url')
                # Use display name from tasks.json instead of raw dataset name
                display_name = filtered_tasks[dataset]['display_name']
                # Round non-null values to 2 decimal places and make clickable if log_url exists
                if value is not None:
                    value = round(value*100, 2)
                    if log_url:
                        value = make_clickable_field(value, log_url)
                row[display_name] = value
        data.append(row)

    results_df = pd.DataFrame(data)
    
    # Round all numeric columns to 2 decimal places
    numeric_cols = results_df.select_dtypes(include=['float64', 'float32']).columns
    results_df[numeric_cols] = results_df[numeric_cols].round(2)

    # Fill null values with "-"
    results_df = results_df.fillna("--")

    if type == "agentic":
        # Include agent column as second column after Model
        results_df.insert(1, 'Agent', make_clickable_field('Basic Agent', 'https://inspect.ai-safety-institute.org.uk/agents.html#sec-basic-agent'))
    
    return gr.components.Dataframe(
        value=results_df,
        datatype=["html" for _ in results_df.columns],
        column_widths=["250px" if c == "Model" else "150px" for c in results_df.columns],
        wrap=False,
    )


black_logo_path = "src/assets/logo-icon-black.png"
white_logo_path = "src/assets/logo-icon-white.png"

demo = gr.Blocks(
    css=custom_css,
    js=custom_js,
    theme=gr.themes.Default(primary_hue=gr.themes.colors.pink),
    fill_height=True,
    fill_width=True,
)
with demo:
    gr.HTML(f"""
    <div id="page-header">
        <div id="header-container">
            <div id="left-container">
                <img id="black-logo" src="/gradio_api/file={black_logo_path}">
                <img id="white-logo" src="/gradio_api/file={white_logo_path}">
            </div>
            <div id="centre-container">
                <h1 style="margin-bottom: 0.25rem;">{TITLE}</h1>
                <p style="color:#eb088a; margin:0; font-size:1.2rem;">Explore Interactive Results &amp; Traces</p>
            </div>
            <div id="right-container">
            </div>
        </div>
    </div>
    """)
    gr.Markdown(INTRODUCTION_TEXT, elem_classes="intro-text", sanitize_html=False)

    with gr.Tabs(elem_classes=["leaderboard-table", "tab-buttons"]) as tabs:
        with gr.TabItem("Base Benchmarks", elem_classes="llm-benchmark-tab-table", id=0):
            build_leaderboard("base")

        with gr.TabItem("Agentic Benchmarks", elem_classes="llm-benchmark-tab-table", id=1):
            build_leaderboard("agentic")

        with gr.TabItem("About", elem_classes="llm-benchmark-tab-table", id=2):
            gr.Markdown(ABOUT_TEXT, elem_classes="markdown-text", sanitize_html=False)

        with gr.TabItem("Reproducibility", elem_classes="llm-benchmark-tab-table", id=3):
            gr.Markdown(REPRODUCIBILITY_TEXT, elem_classes="markdown-text", sanitize_html=False)

assets = [black_logo_path, white_logo_path]
demo.launch(allowed_paths=assets)