feat: Updated record data format

app.py

@@ -46,6 +46,7 @@ def new_comparison(dialog_dict: dict) -> Tuple[str, str, str, str, gr.Row, gr.Ro
 
 def update_scores(score_a: Optional[int],
                   score_b: Optional[int],
+                  win: Optional[int],
                   m1: str,
                   m2: str,
                   d1: str,
@@ -62,12 +63,14 @@ def update_scores(score_a: Optional[int],
         scores["A"] = score_a
     if score_b is not None:
         scores["B"] = score_b
+    if win is not None:
+        scores["win"] = win
 
     # Save the data
-    if "A" in scores and "B" in scores:
+    if "A" in scores and "B" in scores and "win" in scores:
         # if not is_dev:
         with open(result_file_path, "a") as f:
-            f.write(f"{scores['A']}\t{scores['B']}\t{m1}\t{m2}\n")
+            f.write(f"{scores['win']}\t{scores['A']}\t{scores['B']}\t{m1}\t{m2}\n")
 
         new_m1, new_d1, new_m2, new_d2 = sample_pair(dialog_dict)
         return (
@@ -144,6 +147,8 @@ with gr.Blocks(title="1:1 Outpatient Model Simulation Arena", css=css) as demo:
     gr.Markdown("# 🤖 Model Arena Evaluation")
     gr.Markdown("### Compare two model simulations and choose the better one!")
     gr.Markdown("### This scenario assumes that the patient called the hospital's administrative office for an outpatient inquiry.")
+    gr.Markdown("* First step: Arena! Choose the better one between the two simulations.")
+    gr.Markdown("* Second step: Rate the score! Rate both simulations on a scale of 1 to 5. If the two simulations are similar, you may give them the same score regardless of your first choice.")
     gr.Markdown("<br><br>")
     gr.Markdown("---")
 
@@ -189,28 +194,28 @@ with gr.Blocks(title="1:1 Outpatient Model Simulation Arena", css=css) as demo:
     )
 
     # First step: Arena, Choose the only one!
+    score_state = gr.State({})
     vote1.click(
-        fn=lambda: [gr.update(visible=True), gr.update(visible=True), gr.update(visible=False), gr.update(visible=False), gr.update(value="🕐 Please rate both models!")],
-        inputs=[],
-        outputs=[scoreA_row, scoreB_row, vote1_row, vote2_row, msg],
+        fn=lambda m1, m2, d1, d2, scores: update_scores(None, None, 'A', m1, m2, d1, d2, dialog_dict, scores, is_dev, result_save_path),
+        inputs=[model1_name, model2_name, dialog1_box, dialog2_box, score_state],
+        outputs=[model1_name, dialog1_box, model2_name, dialog2_box, arena_row, scoreA_row, scoreB_row, vote1_row, vote2_row, msg, score_state],
     )
     vote2.click(
-        fn=lambda: [gr.update(visible=True), gr.update(visible=True), gr.update(visible=False), gr.update(visible=False), gr.update(value="🕐 Please rate both models!")],
-        inputs=[],
-        outputs=[scoreA_row, scoreB_row, vote1_row, vote2_row, msg],
+        fn=lambda m1, m2, d1, d2, scores: update_scores(None, None, 'B', m1, m2, d1, d2, dialog_dict, scores, is_dev, result_save_path),
+        inputs=[model1_name, model2_name, dialog1_box, dialog2_box, score_state],
+        outputs=[model1_name, dialog1_box, model2_name, dialog2_box, arena_row, scoreA_row, scoreB_row, vote1_row, vote2_row, msg, score_state],
     )
 
     # Second step: Rate, Rate the each score!
-    score_state = gr.State({})
     for btn in scoreA_buttons:
         btn.click(
-            fn=lambda score, m1, m2, d1, d2, scores: update_scores(int(score), None, m1, m2, d1, d2, dialog_dict, scores, is_dev, result_save_path),
+            fn=lambda score, m1, m2, d1, d2, scores: update_scores(int(score), None, None, m1, m2, d1, d2, dialog_dict, scores, is_dev, result_save_path),
             inputs=[gr.State(btn.value), model1_name, model2_name, dialog1_box, dialog2_box, score_state],
             outputs=[model1_name, dialog1_box, model2_name, dialog2_box, arena_row, scoreA_row, scoreB_row, vote1_row, vote2_row, msg, score_state],
         )
     for btn in scoreB_buttons:
         btn.click(
-            fn=lambda score, m1, m2, d1, d2, scores: update_scores(None, int(score), m1, m2, d1, d2, dialog_dict, scores, is_dev, result_save_path),
+            fn=lambda score, m1, m2, d1, d2, scores: update_scores(None, int(score), None, m1, m2, d1, d2, dialog_dict, scores, is_dev, result_save_path),
             inputs=[gr.State(btn.value), model1_name, model2_name, dialog1_box, dialog2_box, score_state],
             outputs=[model1_name, dialog1_box, model2_name, dialog2_box, arena_row, scoreA_row, scoreB_row, vote1_row, vote2_row, msg, score_state],
         )