Upload app.py

app.py

@@ -246,6 +246,43 @@ def evaluate_prompt(prompt: str, dataset_name: str, split: str, num_samples: int
         }
 
 
+def collect_prompt_history(output_dir: str) -> List[Dict]:
+    """
+    Collect all prompts discovered during evolution with their scores.
+
+    Returns a list of dicts with: {prompt, score, iteration, id}
+    """
+    try:
+        prompts = []
+
+        # Look for all program files
+        program_files = sorted(glob.glob(os.path.join(output_dir, "program_*.txt")))
+
+        # Also check for logs to get scores
+        log_dir = os.path.join(output_dir, "logs")
+
+        for pfile in program_files:
+            try:
+                with open(pfile, 'r') as f:
+                    prompt_content = f.read()
+
+                # Extract program ID from filename
+                prog_id = os.path.basename(pfile).replace("program_", "").replace(".txt", "")
+
+                prompts.append({
+                    "prompt": prompt_content,
+                    "id": prog_id,
+                    "file": pfile
+                })
+            except:
+                continue
+
+        return prompts
+    except Exception as e:
+        print(f"Error collecting prompt history: {e}")
+        return []
+
+
 def parse_evolution_history(output_dir: str) -> str:
     """
     Parse evolution history from OpenEvolve output directory.
@@ -522,6 +559,7 @@ def create_config_file(model: str, work_dir: str):
             "elite_ratio": 0.1,
             "explore_ratio": 0.3,
             "exploit_ratio": 0.6,
+            "parallel": False,  # Disable parallel processing to avoid signal errors
         },
         "evaluation": {
             "timeout": None,  # Disable timeout to avoid signal handling issues
@@ -538,7 +576,7 @@ def create_config_file(model: str, work_dir: str):
 
 def optimize_prompt(initial_prompt: str, dataset_name: str, dataset_split: str,
                    model: str, input_field: str, target_field: str,
-                   progress=gr.Progress()) -> Tuple[str, str, str, str]:
+                   progress=gr.Progress()) -> Tuple[str, str, str, str, List[str], int, int]:
     """Run OpenEvolve to optimize the prompt."""
 
     progress(0, desc="Validating inputs...")
@@ -549,7 +587,7 @@ def optimize_prompt(initial_prompt: str, dataset_name: str, dataset_split: str,
     )
 
     if not is_valid:
-        return f"## Validation Failed\n\n{validation_message}", "", "", ""
+        return f"## Validation Failed\n\n{validation_message}", "", "", "", [], 0, 0
 
     progress(0.05, desc=f"Validation passed: {validation_message}")
 
@@ -579,10 +617,10 @@ def optimize_prompt(initial_prompt: str, dataset_name: str, dataset_split: str,
         )
 
         if "error" in initial_eval:
-            return f"## Error\n\n❌ Initial evaluation failed: {initial_eval['error']}", "", "", ""
+            return f"## Error\n\n❌ Initial evaluation failed: {initial_eval['error']}", "", "", "", [initial_prompt], 0, 1
 
         if initial_eval["total"] == 0:
-            return f"## Error\n\n❌ Initial evaluation failed: No samples could be evaluated. This usually means:\n- API key is invalid or has no credits\n- Model is unavailable or rate-limited\n- Dataset fields are incorrect\n- Network connectivity issues\n\nPlease check your configuration and try again.", "", "", ""
+            return f"## Error\n\n❌ Initial evaluation failed: No samples could be evaluated. This usually means:\n- API key is invalid or has no credits\n- Model is unavailable or rate-limited\n- Dataset fields are incorrect\n- Network connectivity issues\n\nPlease check your configuration and try again.", "", "", "", [initial_prompt], 0, 1
 
         initial_results = f"""
 ### Initial Prompt Evaluation
@@ -681,17 +719,25 @@ def optimize_prompt(initial_prompt: str, dataset_name: str, dataset_split: str,
 
             progress(1.0, desc="Complete!")
 
-            return summary, initial_results, evolution_viz, final_results
+            # Collect all discovered prompts for browsing
+            all_prompts = [initial_prompt]  # Start with initial prompt
+            prompt_history = collect_prompt_history(output_dir)
+            for p in prompt_history:
+                all_prompts.append(p["prompt"])
+
+            # Ensure we have the best prompt at the end
+            if best_prompt not in all_prompts:
+                all_prompts.append(best_prompt)
+
+            return summary, initial_results, evolution_viz, final_results, all_prompts, 0, len(all_prompts)
 
         except Exception as e:
-            return f"## Error During Evolution\n\n❌ {str(e)}", initial_results, "", ""
+            return f"## Error During Evolution\n\n❌ {str(e)}", initial_results, "", "", [initial_prompt], 0, 1
 
     finally:
-        # Clean up
-        try:
-            shutil.rmtree(work_dir)
-        except:
-            pass
+        # Don't clean up - keep prompts for browsing
+        # User can manually clean /tmp if needed
+        pass
 
 
 # Create Gradio interface
@@ -784,6 +830,30 @@ with gr.Blocks(title="OpenEvolve Prompt Optimizer", theme=gr.themes.Soft()) as d
         with gr.Column():
             evolution_progress = gr.Markdown("### Evolution Progress\nEvolution progress will appear here during optimization...", visible=True)
 
+    # Prompt History Browser
+    gr.Markdown("---")
+    gr.Markdown("## 📜 Prompt History Browser")
+    gr.Markdown("Browse through all prompts discovered during evolution (initial → intermediate → final)")
+
+    with gr.Row():
+        with gr.Column(scale=8):
+            prompt_display = gr.TextArea(
+                label="",
+                lines=10,
+                interactive=False,
+                placeholder="Prompts will appear here after optimization completes...",
+                show_label=False
+            )
+        with gr.Column(scale=2):
+            prompt_counter = gr.Markdown("**Prompt**: -/-")
+            prev_btn = gr.Button("⬅️ Previous", size="sm")
+            next_btn = gr.Button("Next ➡️", size="sm")
+            gr.Markdown("**Prompt Types:**\n- First = Initial\n- Middle = Intermediate\n- Last = Final Best")
+
+    # Hidden state to store prompt history and current index
+    prompt_history_state = gr.State([])
+    current_prompt_index = gr.State(0)
+
     # Documentation section - in collapsible accordion
     gr.Markdown("---")
     with gr.Accordion("📚 Documentation & Examples", open=False):
@@ -825,11 +895,73 @@ with gr.Blocks(title="OpenEvolve Prompt Optimizer", theme=gr.themes.Soft()) as d
         - [Documentation](https://github.com/algorithmicsuperintelligence/openevolve#readme)
         """)
 
-    optimize_btn.click(
+    # Navigation functions for prompt browser
+    def show_previous_prompt(prompts, current_idx):
+        if not prompts or len(prompts) == 0:
+            return "", "**Prompt**: -/-", 0
+        new_idx = max(0, current_idx - 1)
+        counter_text = f"**Prompt**: {new_idx + 1}/{len(prompts)}"
+        if new_idx == 0:
+            counter_text += " (Initial)"
+        elif new_idx == len(prompts) - 1:
+            counter_text += " (Final Best)"
+        else:
+            counter_text += " (Intermediate)"
+        return prompts[new_idx], counter_text, new_idx
+
+    def show_next_prompt(prompts, current_idx):
+        if not prompts or len(prompts) == 0:
+            return "", "**Prompt**: -/-", 0
+        new_idx = min(len(prompts) - 1, current_idx + 1)
+        counter_text = f"**Prompt**: {new_idx + 1}/{len(prompts)}"
+        if new_idx == 0:
+            counter_text += " (Initial)"
+        elif new_idx == len(prompts) - 1:
+            counter_text += " (Final Best)"
+        else:
+            counter_text += " (Intermediate)"
+        return prompts[new_idx], counter_text, new_idx
+
+    def update_prompt_display(prompts, idx, total):
+        if not prompts or len(prompts) == 0:
+            return "", "**Prompt**: -/-"
+        idx = min(idx, len(prompts) - 1)
+        counter_text = f"**Prompt**: {idx + 1}/{len(prompts)}"
+        if idx == 0:
+            counter_text += " (Initial)"
+        elif idx == len(prompts) - 1:
+            counter_text += " (Final Best)"
+        else:
+            counter_text += " (Intermediate)"
+        return prompts[idx], counter_text
+
+    # Wire up the optimize button
+    optimize_result = optimize_btn.click(
         fn=optimize_prompt,
         inputs=[initial_prompt, dataset_name, dataset_split, model,
                 input_field, target_field],
-        outputs=[summary, initial_results, evolution_progress, final_results]
+        outputs=[summary, initial_results, evolution_progress, final_results,
+                 prompt_history_state, current_prompt_index, gr.State()]  # dummy for total
+    )
+
+    # Update prompt display when optimization completes
+    optimize_result.then(
+        fn=update_prompt_display,
+        inputs=[prompt_history_state, current_prompt_index, gr.State()],
+        outputs=[prompt_display, prompt_counter]
+    )
+
+    # Wire up navigation buttons
+    prev_btn.click(
+        fn=show_previous_prompt,
+        inputs=[prompt_history_state, current_prompt_index],
+        outputs=[prompt_display, prompt_counter, current_prompt_index]
+    )
+
+    next_btn.click(
+        fn=show_next_prompt,
+        inputs=[prompt_history_state, current_prompt_index],
+        outputs=[prompt_display, prompt_counter, current_prompt_index]
     )
 
 if __name__ == "__main__":