pages/long_context.py

import streamlit as st
import pandas as pd
from utils.style import style_long_context

@st.cache_data
def load_data():
    df = pd.read_csv("data/long_context.csv")
    df.dropna(inplace=True) # Drop rows with any missing values
    return df

def show():
    st.title("Long Context Leaderboard")

    # Load and style data
    df = load_data()
    styled_df = style_long_context(df)

    # Display the dataframe with built-in sort on column click
    st.dataframe(
        styled_df,
        use_container_width=True,
        height=35*(len(df)+1),
        hide_index=True,
        column_config={
            "Model": st.column_config.TextColumn(width="large"),
            "8K": st.column_config.NumberColumn(format="%.2f"),
            "16K": st.column_config.NumberColumn(format="%.2f"),
            "32K": st.column_config.NumberColumn(format="%.2f"),
            "Average↑": st.column_config.NumberColumn(
                format="%.2f",
                help="Average across all context lengths"
            )
        }
    )

    # Optionally, keep some explanatory text
    st.markdown("""
    **Context Lengths**:
    - 8K: 8,000 tokens
    - 16K: 16,000 tokens
    - 32K: 32,000 tokens
                
    **Colors**:
    - Yellow: reasoning model
    - Green: linear attention hybrid model
    - Blue: SSM hybrid model

    **Benchmark Details**:
    - Evaluated on Symbolic, Medium, and Hard subtasks.
    - Area Under Curve(AUC) Metrics is Used to Compare between LLM Performance.
    - AUC is calculated using np.trapz function.
    - AUC scores aggregated across context lengths.
    - Larger context evaluations limited by compute constraints and model performance.
    """)