Rewrote interruption & improved UI (#9)

- Rewrote interruption & improved UI (23f15c31c502d9e13b24caa75169353cae64aa1f)


Co-authored-by: Kai <kai-aizip@users.noreply.huggingface.co>

app.py

@@ -3,15 +3,12 @@ import random
 import pandas as pd
 import os
 import threading
-from threading import Event
 from utils.data_loader import get_random_example
 from utils.models import generate_summaries, model_names
 from utils.ui_helpers import toggle_context_display, update_feedback, get_context_html
 from utils.leaderboard import load_leaderboard_data, submit_vote_with_elo, generate_leaderboard_html
 from utils.vote_logger import save_vote_details
-
-# Global interrupt mechanism for model generation
-generation_interrupt = Event()
+from utils.shared import generation_interrupt  # Import from shared module
 
 # Feedback options for different voting outcomes
 feedback_options = {
@@ -251,7 +248,7 @@ with gr.Blocks(theme=gr.themes.Default(
                             gr.Markdown("### Model B was actually:", elem_classes="section-heading")
                             model_b_reveal = gr.Markdown("", elem_classes="model-reveal model-b-reveal")
                      
-                    gr.HTML("<div style='height: 10px;'></div>")
+                    gr.HTML("<hr>")
                     
                     # Try another button
                     with gr.Row(elem_classes=["control-buttons"]):
@@ -266,7 +263,6 @@ with gr.Blocks(theme=gr.themes.Default(
                 gr.Markdown("""### About Elo Ratings
                 
 The Elo rating system provides a more accurate ranking than simple win rates:
-
 - All models start at 1500 points
 - Points are exchanged after each comparison based on the expected outcome
 - Beating a stronger model earns more points than beating a weaker one
@@ -289,6 +285,17 @@ The Elo rating system provides a more accurate ranking than simple win rates:
             False
         ]
 
+    # Add this new function to show a loading state
+    def show_loading_state():
+        return [
+            gr.update(value="Loading new question and summaries...", interactive=False),
+            gr.update(value="Loading new question and summaries...", interactive=False),
+            gr.update(interactive=False),
+            gr.update(interactive=False),
+            gr.update(interactive=False),
+            gr.update(interactive=False)
+        ]
+
     # Event handling
     # Toggle context display
     context_toggle_btn.click(
@@ -323,6 +330,11 @@ The Elo rating system provides a more accurate ranking than simple win rates:
     # Random Question and Try Another buttons with interruption
     for btn in [random_question_btn, try_another_btn]:
         btn.click(
+            fn=show_loading_state,  # First show loading state
+            inputs=[],
+            outputs=[summary_a_display, summary_b_display, vote_button_a, 
+                    vote_button_b, vote_button_tie, vote_button_neither]
+        ).then(
             fn=handle_new_example_click,
             inputs=[],
             outputs=[current_example]