slow down api call

app.py

@@ -140,16 +140,25 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             }
 
     import concurrent.futures
-    # Max workers = 10 -> Process 10 questions at the same time
-    with concurrent.futures.ThreadPoolExecutor(max_workers=10) as executor:
-        # Submit all items
-        futures = {executor.submit(process_item, item): item for item in questions_data}
+    import time
+    
+    # Max workers = 2 -> Groq API has strict Token and Request Per Minute limits.
+    # 2 workers with a slight stagger prevents immediate bursting.
+    with concurrent.futures.ThreadPoolExecutor(max_workers=2) as executor:
+        futures = {}
+        for item in questions_data:
+            futures[executor.submit(process_item, item)] = item
+            time.sleep(3) # Stagger starting requests by 3 seconds to avoid bursting Rate Limits
+
         for future in concurrent.futures.as_completed(futures):
             res = future.result()
             if res:
                 answers_payload.append({"task_id": res["task_id"], "submitted_answer": res["submitted_answer"]})
                 results_log.append({"Task ID": res["task_id"], "Question": res["question"], "Submitted Answer": res["submitted_answer"]})
 
+                # Additional delay after finishing a question to let Token bucket refill
+                time.sleep(2)
+
     if not answers_payload:
         print("Agent did not produce any answers to submit.")
         return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)