test

agent.py

@@ -523,32 +523,15 @@ CRITICAL RULES:
         if isinstance(msg, HumanMessage) and "[Attached File Local Path:" in msg.content:
             messages.append(HumanMessage(content="IMPORTANT: I see an image path in the message. I MUST call the analyze_image tool IMMEDIATELY in my next step to see it."))
 
-    # Multi-step ReAct Loop (Up to 12 reasoning steps)
-    max_steps = 12
+    # One-shot reasoning for better latency and more consistent accuracy on short QA.
     draft_response = None
     current_tier = 0
-    
-    for step in range(max_steps):
-        if step > 0:
-            # Prevents Groq API Request/Tokens Per Minute exceptions when deep reasoning
-            time.sleep(4)
-            
-        print(f"--- ReAct Step {step + 1} ---")
-        
-        # Max history truncation to avoid 413 Request Too Large errors
-        # Keep SystemMessage, first HumanMessage, and the last 6 messages
-        safe_messages = messages[:2] + messages[-6:] if len(messages) > 10 else messages
-        
-        ai_msg, current_tier = smart_invoke(safe_messages, use_tools=True, start_tier=current_tier)
-        messages.append(ai_msg)
-
-        # Check if the model requested tools
-        tool_calls = getattr(ai_msg, "tool_calls", None) or []
-        if not tool_calls:
-            # Model decided it has enough info to answer
-            draft_response = ai_msg
-            print(f"Model found answer or stopped tools: {ai_msg.content}")
-            break
+
+    print("--- One-shot response invocation ---")
+    ai_msg, current_tier = smart_invoke(messages, use_tools=False, start_tier=current_tier)
+    messages.append(ai_msg)
+    draft_response = ai_msg
+    print(f"Model returned answer: {ai_msg.content}")
 
         # Execute requested tools and append their text output into the conversation
         for tool_call in tool_calls:

app.py

@@ -143,13 +143,10 @@ def run_and_submit_all(profile: Optional[gr.OAuthProfile] = None):
     import concurrent.futures
     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
+    # Improve throughput while respecting rate limits; avoid fixed sleep delays that slow down the entire run.
+    max_workers = min(8, len(questions_data)) if questions_data else 1
+    with concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) as executor:
+        futures = {executor.submit(process_item, item): item for item in questions_data}
 
         for future in concurrent.futures.as_completed(futures):
             res = future.result()
@@ -157,9 +154,6 @@ def run_and_submit_all(profile: Optional[gr.OAuthProfile] = None):
                 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)