src/json_leaderboard.py

import json
import pandas as pd
from pathlib import Path


SIZE_BANDS = ["All", "<10B", "10-32B", "32-100B", ">100B", "Unknown"]
MODALITIES = ["All", "VL", "Omni"]


def load_leaderboard_from_json(json_path="leaderboard_data.json"):
    """Load leaderboard data from JSON file"""
    try:
        with open(json_path, 'r', encoding='utf-8') as f:
            data = json.load(f)
        return data['leaderboard']
    except FileNotFoundError:
        print(f"JSON file {json_path} not found")
        return []
    except json.JSONDecodeError:
        print(f"Error decoding JSON file {json_path}")
        return []


def _in_size_band(params_b, band):
    if band == "All":
        return True
    if params_b is None or pd.isna(params_b):
        return band == "Unknown"
    if band == "<10B":
        return params_b < 10
    if band == "10-32B":
        return 10 <= params_b < 32
    if band == "32-100B":
        return 32 <= params_b < 100
    if band == ">100B":
        return params_b >= 100
    if band == "Unknown":
        return False
    return True


def create_leaderboard_df(
    json_path="leaderboard_data.json",
    open_source_only=False,
    size_band="All",
    modality="All",
):
    """Create a pandas DataFrame from JSON leaderboard data, with optional filters."""
    leaderboard_data = load_leaderboard_from_json(json_path)

    if not leaderboard_data:
        return pd.DataFrame()

    df = pd.DataFrame(leaderboard_data)

    # Backfill optional columns for older JSON entries
    if 'params_b' not in df.columns:
        df['params_b'] = None
    if 'modality' not in df.columns:
        df['modality'] = "VL"

    # Apply filters
    if open_source_only:
        df = df[df['open_source'] == True]
    if modality and modality != "All":
        df = df[df['modality'] == modality]
    if size_band and size_band != "All":
        df = df[df['params_b'].apply(lambda v: _in_size_band(v, size_band))]

    if df.empty:
        return pd.DataFrame()

    # Sort by ACC score (descending) so medal icons reflect the filtered view
    df = df.sort_values('acc', ascending=False).reset_index(drop=True)

    def add_ranking_icon_and_link(index, model_name, paper_link):
        if index == 0:
            return f'🥇 <a href="{paper_link}" target="_blank">{model_name}</a>'
        elif index == 1:
            return f'🥈 <a href="{paper_link}" target="_blank">{model_name}</a>'
        elif index == 2:
            return f'🥉 <a href="{paper_link}" target="_blank">{model_name}</a>'
        else:
            return f'<a href="{paper_link}" target="_blank">{model_name}</a>'

    display_df = pd.DataFrame({
        'Model': [add_ranking_icon_and_link(i, model, link) for i, (model, link) in enumerate(zip(df['model'], df['link']))],
        'Release Date': df['release_date'],
        'HF Model': df['hf'].apply(lambda x: f'<a href="{x}" target="_blank">🤗</a>' if x != "-" else "-"),
        'Modality': df['modality'],
        'Parameters': df['params'],
        'Open Source': df['open_source'].apply(lambda x: '✓' if x else '✗'),
        'ACC Score': df['acc'].apply(lambda x: f"{x:.1f}")
    })

    return display_df


def get_leaderboard_stats(json_path="leaderboard_data.json"):
    """Get statistics about the leaderboard"""
    leaderboard_data = load_leaderboard_from_json(json_path)

    if not leaderboard_data:
        return {}

    df = pd.DataFrame(leaderboard_data)

    stats = {
        'total_models': len(df),
        'open_source_models': int(df['open_source'].sum()),
        'moe_models': int(df['moe'].apply(lambda x: 1 if x is True else 0).sum()),
        'avg_acc': df['acc'].mean(),
        'max_acc': df['acc'].max(),
        'min_acc': df['acc'].min()
    }

    return stats