release full benchmark viewer

pages/benchmark_viewer.py

@@ -0,0 +1,99 @@
+import streamlit as st
+import pandas as pd
+import plotly.graph_objects as go
+import numpy as np
+
+def show():
+    st.title("Benchmark Results Dashboard")
+
+    @st.cache_data
+    def load_data():
+        """Load and process benchmark results, handling zero accuracy."""
+        try:
+            df = pd.read_csv('data/processed_results.csv')
+        except FileNotFoundError:
+            st.error("File 'processed_results.csv' not found.")
+            st.stop()
+
+        epsilon = 1e-6
+        num_zero_acc = (df['accuracy'] <= 0).sum()
+        if num_zero_acc > 0:
+            st.warning(f"Found {num_zero_acc} zero/negative accuracy values. Replacing with {epsilon}.")
+            df.loc[df['accuracy'] <= 0, 'accuracy'] = epsilon
+
+        df['log_accuracy'] = np.log(df['accuracy'])
+        return df
+
+    df = load_data()
+
+    # Filters
+    st.header("Filters")
+
+    col1, col2, col3 = st.columns(3)
+
+    with col1:
+        datasets = df['dataset'].unique()
+        selected_datasets = st.multiselect("Dataset(s)", datasets, default=datasets)
+        # Filter data based on selected datasets first
+        filtered_df = df[df['dataset'].isin(selected_datasets)]
+
+        lengths = sorted(filtered_df['length'].unique())
+        # Disable length filter if no datasets are selected
+        disabled = not selected_datasets
+        selected_lengths = st.multiselect("Length(s)", lengths, default=lengths if not disabled and lengths else [], disabled=disabled)
+
+
+    with col2:
+        # Single Model Multiselect (filtered by selected datasets)
+        available_models = filtered_df['model'].unique()
+        selected_models = st.multiselect("Model(s)", available_models, default=available_models) # Handle empty defaults
+
+    with col3:
+        min_op, max_op = st.slider("Op Range", int(filtered_df['op'].min()), int(filtered_df['op'].max()), (int(filtered_df['op'].min()), int(filtered_df['op'].max())))
+        min_acc, max_acc = st.slider("Accuracy Range", float(filtered_df['accuracy'].min()), float(filtered_df['accuracy'].max()), (float(filtered_df['accuracy'].min()), float(filtered_df['accuracy'].max())))
+
+
+    filtered_df = filtered_df[filtered_df['model'].isin(selected_models) & filtered_df['length'].isin(selected_lengths)]
+
+
+    filtered_df = filtered_df[(filtered_df['op'] >= min_op) & (filtered_df['op'] <= max_op) & (filtered_df['accuracy'] >= min_acc) & (filtered_df['accuracy'] <= max_acc)]
+
+
+    if filtered_df.empty:
+        st.warning("No data for selected filters.")
+        st.stop()
+
+    def plot_data(filtered_df, selected_models, selected_lengths, log_scale=False):
+        """Plots data vs N, showing different datasets for the same model."""
+        fig = go.Figure()
+
+        for model in selected_models:
+            for length in selected_lengths:
+                for dataset in filtered_df['dataset'].unique():
+                    subset_df = filtered_df[(filtered_df['model'] == model) & (filtered_df['length'] == length) & (filtered_df['dataset'] == dataset)]
+                    if not subset_df.empty:
+                        y_data = subset_df['log_accuracy'] if log_scale else subset_df['accuracy']
+                        fig.add_trace(go.Scatter(
+                            x=subset_df['op'],
+                            y=y_data,
+                            mode='lines+markers',
+                            name=f'{model} Length {length} ({dataset})',
+                            marker=dict(size=6)
+                        ))
+
+        y_title = "Log(Accuracy)" if log_scale else "Accuracy"
+        fig.update_layout(title=f"{y_title} vs Op", xaxis_title="Op", yaxis_title=y_title)
+        return fig
+
+    view_option = st.radio("View", ["Accuracy", "Log(Accuracy)"])
+
+    if view_option == "Accuracy":
+        fig = plot_data(filtered_df, selected_models, selected_lengths, log_scale=False)
+    else:  # Log(Accuracy)
+        fig = plot_data(filtered_df, selected_models, selected_lengths, log_scale=True)
+
+    st.plotly_chart(fig, use_container_width=True)
+
+    if st.checkbox("Show Data Table"):
+        st.subheader("Filtered Data")
+        st.write(filtered_df)

pages/long_context.py

@@ -41,6 +41,5 @@ def show():
     **Benchmark Details**:
     - Evaluated on Symbolic, Medium, and Hard subtasks
     - AUC scores aggregated across context lengths
-    - Larger context evaluations limited by compute constraints
-    - Scores normalized across task complexities
+    - Larger context evaluations limited by compute constraints and model performance
     """)

pages/zero_context.py

@@ -40,7 +40,8 @@ def show():
 
     # You can leave your explanation/description below
     st.markdown("""
-    **Evaluation Criteria:**
-    - **AUC Calculation:** ...
-    - **Threshold Ops:** ...
+    **Benchmark Details**:
+    - Evaluated on Symbolic, Medium, and Hard subtasks.
+    - Area Under Curve Metrics is Used to Compare between LLM Performance.
+    - AUC is calculated using np.trapz function.
     """)