Update app.py

app.py

@@ -1,9 +1,6 @@
 # Hey, stranger! this code is for use of free rate of gemini llm 
-# which is limited by RPM (15/30). Testing has shown that if I put 
-# request delay 10 then search drops out timed out.
-# Nevertheless, it scrored 35% which is good for me while two questions 
-# were dropped due to exceeding RPM. So, it is still possible to improve,
-# e.g. deploying gemini 2.0 flash lite which has double RPM limit.
+# which is limited by RPM (15/30). 
+# Nevertheless, it scrored 35% which is good for me...
 # Try it out!
 
 import os
@@ -23,7 +20,7 @@ from token_bucket import Limiter, MemoryStorage
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 # Rate limiting configuration
-MAX_MODEL_CALLS_PER_MINUTE = 10  # Conservative buffer below 15 RPM
+MAX_MODEL_CALLS_PER_MINUTE = 13  # Conservative buffer below 15 RPM
 RATE_LIMIT = MAX_MODEL_CALLS_PER_MINUTE
 TOKEN_BUCKET_CAPACITY = RATE_LIMIT
 TOKEN_BUCKET_REFILL_RATE = RATE_LIMIT / 60.0  # Tokens per second
@@ -73,7 +70,7 @@ async def process_question(agent, question_text: str, task_id: str, results_log:
     """Process a single question with global rate limiting."""
     submitted_answer = None
     max_retries = 3
-    retry_delay = 6  # 6 seconds for 10 RPM
+    retry_delay = 5  # 6 seconds for 10 RPM
 
     for attempt in range(max_retries):
         try:
@@ -84,7 +81,7 @@ async def process_question(agent, question_text: str, task_id: str, results_log:
             print(f"Processing task {task_id} (attempt {attempt + 1})...")
             submitted_answer = await asyncio.wait_for(
                 agent(question_text, task_id),
-                timeout=60  # 60-second timeout per question
+                timeout=35  # 60-second timeout per question
             )
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
             print(f"Completed task {task_id} with answer: {submitted_answer[:50]}...")