Update app.py

app.py

@@ -1,7 +1,9 @@
 import gradio as gr
 import json
 import os
+import time
 import requests
+from tools import TOOLS
 import traceback
 from agent import solve_task, load_tasks
 from mistral_hf_wrapper import MistralInference
@@ -11,34 +13,55 @@ API_TOKEN = os.getenv("HF_TOKEN")
 USERNAME = os.getenv("HF_USERNAME")
 CODE_LINK = os.getenv("HF_CODE_LINK")
 
+MAX_RETRIES = 3
+INFERENCE_TIMEOUT = 45  # seconds
+
 def run_and_submit_all():
     model = MistralInference(api_url=API_URL, api_token=API_TOKEN)
     tasks = load_tasks()
 
+    print(f"[INFO] Loaded {len(tasks)} tasks from metadata.jsonl")
+
     if not tasks:
         return "No tasks loaded from metadata.jsonl. Make sure the file exists and is valid."
 
-    print(f"[INFO] Loaded {len(tasks)} tasks from metadata.jsonl")
-
     answers = []
     for i, task in enumerate(tasks):
-        print(f"[INFO] Solving task {i+1}/{len(tasks)}: {task.get('task_id', 'UNKNOWN')}")
-        try:
-            result = solve_task(task, model)
-            print(f"[INFO]Answer: {result['submitted_answer'][:100]}...")
-            if not result.get("submitted_answer"):
-                result["submitted_answer"] = "ERROR: Empty model response"
-            answers.append(result)
-        except Exception as e:
-            tb = traceback.format_exc()
-            print(f"[ERROR] Task failed: {e}\n{tb}")
-            answers.append({
-                "question_id": task.get("task_id", "UNKNOWN"),
-                "submitted_answer": f"ERROR: {str(e)}"
-            })
+        task_id = task.get("task_id", f"UNKNOWN-{i}")
+        print(f"[INFO] Solving task {i+1}/{len(tasks)}: {task_id}")
+        
+        attempt = 0
+        success = False
+        while attempt < MAX_RETRIES and not success:
+            attempt += 1
+            try:
+                start = time.time()
+                result = solve_task(task, tools=TOOLS)
+                duration = time.time() - start
+
+                # Handle empty responses
+                if not result.get("submitted_answer"):
+                    raise ValueError("Empty model response")
 
-    if not answers:
-        return "No answers generated. Check model response or network."
+                print(f"[INFO] Answer in {duration:.1f}s: {result['submitted_answer'][:100]}...")
+                answers.append({
+                    "task_id": task_id,
+                    "submitted_answer": result["submitted_answer"].strip()
+                })
+                success = True
+
+            except Exception as e:
+                print(f"[WARN] Attempt {attempt} failed: {e}")
+                if attempt == MAX_RETRIES:
+                    answers.append({
+                        "task_id": task_id,
+                        "submitted_answer": f"ERROR: {str(e)}"
+                    })
+
+    with open("partial_answers.json", "w", encoding="utf-8") as f:
+        json.dump(answers, f, indent=2)
+
+    print("[INFO] Submitting answers to leaderboard...")
 
     try:
         res = requests.post(
@@ -49,16 +72,19 @@ def run_and_submit_all():
                 "agent_code": CODE_LINK,
                 "answers": answers
             },
-            timeout=60  # increase timeout here too
+            timeout=60  # Set submission timeout
         )
-
         if res.ok:
+            print("[INFO]  Submission successful")
             return json.dumps(res.json(), indent=2)
         else:
             return f"Error submitting: {res.status_code} - {res.text}"
 
+    except requests.exceptions.Timeout:
+        return " Submission timed out. Try again later."
     except Exception as e:
-        return f"[SUBMISSION ERROR] {str(e)}"
+        return f" Submission failed with exception: {e}"
+
 # Gradio interface
 gr.Interface(
     fn=run_and_submit_all,