update

app.py

@@ -45,18 +45,22 @@ import random
 random.seed(42)
 np.random.seed(42)
 def sample_an_feedback(selected_models):
-    shuffled_dataset = dataset.shuffle()
     feedback = None
-    for example in tqdm(shuffled_dataset, total=len(shuffled_dataset), desc="Searching for valid examples"):
+    all_valid_feedbacks = []
+    all_valid_model_responses = []
+    for example in tqdm(dataset, total=len(dataset), desc="Searching for valid examples"):
         example_model_responses = example['responses']
         valid_model_responses = [model_response for model_response in example_model_responses if model_response['model'] in selected_models]
         if len(valid_model_responses) >= 2:
-            feedback = example
-            model_response_1, model_response_2 = random.sample(valid_model_responses, 2)
-            break
-    if not feedback:
+            all_valid_feedbacks.append(example)
+            all_valid_model_responses.append(random.sample(valid_model_responses, 2))
+    if len(all_valid_feedbacks) == 0:
         return gr.Exit("No valid examples found. Please select other models.")
     
+    random_idx = random.randint(0, len(all_valid_feedbacks) - 1)
+    feedback = all_valid_feedbacks[random_idx]
+    model_response_1, model_response_2 = all_valid_model_responses[random_idx]
+
     plan_history = {
         "user": [
             feedback['query'],