Upload app.py

app.py

@@ -545,10 +545,10 @@ def optimize_prompt(initial_prompt: str, dataset_name: str, dataset_split: str,
         progress(0.15, desc="Creating configuration...")
         config_path = create_config_file(model, work_dir)
 
-        # Run initial evaluation
-        progress(0.2, desc="Running initial evaluation on 100 samples...")
+        # Run initial evaluation (using 20 samples to save API calls)
+        progress(0.2, desc="Running initial evaluation on 20 samples...")
         initial_eval = evaluate_prompt(
-            initial_prompt, dataset_name, dataset_split, 100,
+            initial_prompt, dataset_name, dataset_split, 20,
             model, input_field, target_field
         )
 
@@ -598,7 +598,7 @@ def optimize_prompt(initial_prompt: str, dataset_name: str, dataset_split: str,
             # Parse evolution history for visualization
             evolution_viz = parse_evolution_history(output_dir)
 
-            progress(0.85, desc="Evaluating best evolved prompt...")
+            progress(0.85, desc="Evaluating best evolved prompt on 20 samples...")
 
             # Get the best prompt
             best_prompt_path = os.path.join(output_dir, "best_program.txt")
@@ -608,9 +608,9 @@ def optimize_prompt(initial_prompt: str, dataset_name: str, dataset_split: str,
             else:
                 best_prompt = initial_prompt
 
-            # Evaluate best prompt
+            # Evaluate best prompt (using 20 samples like initial eval for consistency)
             final_eval = evaluate_prompt(
-                best_prompt, dataset_name, dataset_split, 100,
+                best_prompt, dataset_name, dataset_split, 20,
                 model, input_field, target_field
             )
 
@@ -641,7 +641,8 @@ def optimize_prompt(initial_prompt: str, dataset_name: str, dataset_split: str,
 ### Summary
 - **Dataset**: {dataset_name} ({dataset_split} split)
 - **Model**: {model}
-- **Evaluation**: Staged (20 samples → 100 if score ≥ 0.5)
+- **Initial/Final Eval**: 20 samples each
+- **Evolution Eval**: Staged (20 → 100 if score ≥ 0.5)
 - **Iterations**: 10
 
 ### Results