Update app.py

app.py

@@ -228,10 +228,10 @@ def battle_arena(prompt):
     
     # Check for API errors in responses
     if any("Error: Unable to get response from the model" in msg["content"]
-           for msg in response_a + response_b 
+           for msg in response_a + response_b
            if msg["role"] == "assistant"):
         return (
-            [], [], None, None,
+            [], [], None, None, None, None,
             gr.update(value=[]),
             gr.update(value=[]),
             gr.update(interactive=False, value="Voting Disabled - API Error"),
@@ -247,10 +247,11 @@ def battle_arena(prompt):
     nickname_b = random.choice(config.model_nicknames)
     
     # The responses are already in the correct format, no need to reformat
+    # Randomly swap positions for blind testing
     if random.choice([True, False]):
-        logger.warning(f"NOT SWAPPED: left={model_a}, right={model_b}")
+        logger.warning(f"NOT SWAPPED: left={model_a}({nickname_a}), right={model_b}({nickname_b})")
         return (
-            response_a, response_b, model_a, model_b,
+            response_a, response_b, model_a, model_b, nickname_a, nickname_b,
             gr.update(label=nickname_a, value=response_a),
             gr.update(label=nickname_b, value=response_b),
             gr.update(interactive=True, value=f"Vote for {nickname_a}"),
@@ -262,9 +263,9 @@ def battle_arena(prompt):
             gr.update(value="Ready for your vote! 🗳️", visible=True)
         )
     else:
-        logger.warning(f"SWAPPED: left={model_b}, right={model_a}")
+        logger.warning(f"SWAPPED: left={model_b}({nickname_a}), right={model_a}({nickname_b})")
         return (
-            response_b, response_a, model_b, model_a,
+            response_b, response_a, model_b, model_a, nickname_a, nickname_b,
             gr.update(label=nickname_a, value=response_b),
             gr.update(label=nickname_b, value=response_a),
             gr.update(interactive=True, value=f"Vote for {nickname_a}"),
@@ -276,7 +277,7 @@ def battle_arena(prompt):
             gr.update(value="Ready for your vote! 🗳️", visible=True)
         )
 
-def record_vote(prompt, left_response, right_response, left_model, right_model, choice, battles_display, submit_btn_state):
+def record_vote(prompt, left_response, right_response, left_model, right_model, left_nickname, right_nickname, voted_nickname, battles_display, submit_btn_state):
     # Check if outputs are generated
     if not left_response or not right_response or not left_model or not right_model:
         return (
@@ -291,8 +292,20 @@ def record_vote(prompt, left_response, right_response, left_model, right_model,
             gr.update()
         )
     
-    winner = left_model if choice == "Left is better" else right_model
-    loser = right_model if choice == "Left is better" else left_model
+    # Determine winner based on the voted nickname
+    if voted_nickname == left_nickname:
+        winner = left_model
+        loser = right_model
+        winner_nickname = left_nickname
+        loser_nickname = right_nickname
+    else:
+        winner = right_model
+        loser = left_model
+        winner_nickname = right_nickname
+        loser_nickname = left_nickname
+    
+    logger.warning(f"VOTE: voted_nickname={voted_nickname}, left={left_nickname}({left_model}), right={right_nickname}({right_model})")
+    logger.warning(f"VOTE RESULT: winner={winner_nickname}({winner}), loser={loser_nickname}({loser})")
     
     # Update the leaderboard
     battle_results = update_leaderboard(winner, loser)
@@ -302,9 +315,9 @@ def record_vote(prompt, left_response, right_response, left_model, right_model,
     
     result_message = f"""
 🎉 Vote recorded! You're awesome! 🌟
-🔵 In the left corner: {get_human_readable_name(left_model)}
-🔴 In the right corner: {get_human_readable_name(right_model)}
-🏆 And the champion you picked is... {get_human_readable_name(winner)}! 🥇
+👤 **{left_nickname}** was actually: {get_human_readable_name(left_model)}
+👤 **{right_nickname}** was actually: {get_human_readable_name(right_model)}
+🏆 And the champion you picked is... **{winner_nickname}** ({get_human_readable_name(winner)})! 🥇
     """
     
     return (
@@ -327,8 +340,10 @@ def new_battle():
         "", # Reset prompt_input
         gr.update(value=[], label=nickname_a),  # Reset left Chatbot
         gr.update(value=[], label=nickname_b),  # Reset right Chatbot
-        None,
-        None,
+        None,  # left_model
+        None,  # right_model
+        None,  # left_nickname_state
+        None,  # right_nickname_state
         gr.update(interactive=False, value=f"Vote for {nickname_a}"),
         gr.update(interactive=False, value=f"Vote for {nickname_b}"),
         gr.update(interactive=False, visible=False),  # Reset Tie button
@@ -461,8 +476,11 @@ with gr.Blocks() as demo:
             left_model = gr.Textbox(label="🔵 Left Model", interactive=False)
             right_model = gr.Textbox(label="🔴 Right Model", interactive=False)
         
-        previous_prompt = gr.State("")  # Add this line to store the previous prompt
-        tie_count = gr.State(0)  # Add this line to keep track of tie count
+        # State variables
+        previous_prompt = gr.State("")
+        tie_count = gr.State(0)
+        left_nickname_state = gr.State("")
+        right_nickname_state = gr.State("")
         
         new_battle_btn = gr.Button("New Battle")
     
@@ -497,22 +515,26 @@ with gr.Blocks() as demo:
         battle_arena,
         inputs=prompt_input,
         outputs=[
-            left_output, right_output, left_model, right_model,
+            left_output, right_output, left_model, right_model, left_nickname_state, right_nickname_state,
             left_output, right_output, left_vote_btn, right_vote_btn,
             tie_btn, previous_prompt, tie_count, model_names_row, result
         ]
     )
     
     left_vote_btn.click(
-        lambda *args: record_vote(*args, "Left is better"),
-        inputs=[prompt_input, left_output, right_output, left_model, right_model, battles_counter, submit_btn],
+        lambda prompt, left_resp, right_resp, left_mod, right_mod, left_nick, right_nick, battles, submit: record_vote(
+            prompt, left_resp, right_resp, left_mod, right_mod, left_nick, right_nick, left_nick, battles, submit
+        ),
+        inputs=[prompt_input, left_output, right_output, left_model, right_model, left_nickname_state, right_nickname_state, battles_counter, submit_btn],
         outputs=[result, leaderboard, elo_leaderboard, left_vote_btn,
                  right_vote_btn, tie_btn, model_names_row, battles_counter, submit_btn]
     )
     
     right_vote_btn.click(
-        lambda *args: record_vote(*args, "Right is better"),
-        inputs=[prompt_input, left_output, right_output, left_model, right_model, battles_counter, submit_btn],
+        lambda prompt, left_resp, right_resp, left_mod, right_mod, left_nick, right_nick, battles, submit: record_vote(
+            prompt, left_resp, right_resp, left_mod, right_mod, left_nick, right_nick, right_nick, battles, submit
+        ),
+        inputs=[prompt_input, left_output, right_output, left_model, right_model, left_nickname_state, right_nickname_state, battles_counter, submit_btn],
         outputs=[result, leaderboard, elo_leaderboard, left_vote_btn,
                  right_vote_btn, tie_btn, model_names_row, battles_counter, submit_btn]
     )
@@ -526,7 +548,7 @@ with gr.Blocks() as demo:
     new_battle_btn.click(
         new_battle,
         outputs=[prompt_input, left_output, right_output, left_model,
-                right_model, left_vote_btn, right_vote_btn, tie_btn,
+                right_model, left_nickname_state, right_nickname_state, left_vote_btn, right_vote_btn, tie_btn,
                 result, leaderboard, model_names_row, elo_leaderboard, tie_count, submit_btn]
     )