Fallback agent

Added logic to switch Gemini agent model after RPD has been reached

agent.py

@@ -151,6 +151,8 @@ class GeminiAgent:
     def __init__(self, native_multimodal: bool = True, model_id: str = "gemini/gemini-2.5-flash-lite"):
     # def __init__(self, native_multimodal: bool = True, model_id: str = "gemini/gemini-3-flash-preview"):
         self.native_multimodal = native_multimodal
+        self.model_id = model_id
+
         if self.native_multimodal:
             client = genai.Client(api_key=os.environ.get("GOOGLE_API_KEY"))
 
@@ -178,7 +180,7 @@ class GeminiAgent:
             self.tools = AGENT_TOOLS
         self.gemini_agent = CodeAgent(
             name           = "gemini_agent",
-            description    = "Gemini CodeAgent",
+            description    = f"Gemini CodeAgent ({model_id})",
             model          = self.model,
             tools          = self.tools,
             add_base_tools = True,        # probably redundant, but it does not hurt
@@ -188,7 +190,7 @@ class GeminiAgent:
             max_print_outputs_length=1_000_000
             )
         
-        print("✅ Gemini agent initialized")
+        print(f"✅ Gemini agent initialized with model: {model_id}")
                 
     def __call__(self, question: str, file_path: Optional[str] = None) -> str:
         prompt = f"{self.system_prompt}\n\nQuestion: {question}"

app.py

@@ -8,6 +8,23 @@ from agent import BasicAgent, GeminiAgent
 from typing import Optional
 from litellm.exceptions import RateLimitError, ContextWindowExceededError
 
+# Gemini agent configs
+MAX_TOTAL_REQUESTS = 20
+request_count = 0
+active_model_index = 0
+
+GEMINI_MODELS = [
+    "gemini/gemini-2.5-flash-lite",
+    "gemini/gemini-2.5-flash",
+]
+
+# Helper: (re)load Gemini agent with fallback
+def load_gemini_agent(model_index: int) -> GeminiAgent:
+    model_id = GEMINI_MODELS[model_index]
+    print(f"Loading Gemini agent with model: {model_id}")
+    return GeminiAgent(model_id=model_id)
+
+
 # (Keep Constants as is)
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
@@ -29,6 +46,10 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
     global interrupt_flag 
     interrupt_flag = False
 
+    global request_count, active_model_index
+    request_count = 0
+    active_model_index = 0
+
     # --- Determine HF Space Runtime URL and Repo URL ---
     space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code
 
@@ -45,7 +66,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
 
     # 1. Instantiate Agent (modify this part to create your agent)
     try:
-        agent = GeminiAgent()
+        agent = load_gemini_agent(active_model_index)
         agent_type = "GeminiAgent"
     except Exception as main_agent_error:
         print(f"{agent_type} failed to initialize: {main_agent_error}.")
@@ -55,7 +76,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             print(f"Falling back to {agent_type}.")
         except Exception as secondary_agent_error:
             print(f"{agent_type} failed to initialize: {secondary_agent_error}.")
-            agent_type = "None"
+            agent_type = "None" # replace with BasicAgent() if credits allow
             return f"Error initializing agent: {e}", None
             
     # In the case of an app running as a hugging Face space, this link points toward your codebase ( usefull for others so please keep it public)
@@ -102,7 +123,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             continue
 
         # CONTENT FILTER SKIP (using .lower() for case-insensitivity)
-        filter_keywords = ["chess"]
+        filter_keywords = ["chess", "video"]
         question_words = set(question_text.lower().split()) # Only matches if the exact word is used
         if any(word in question_words for word in filter_keywords):
             print(f"Skipping filtered question: {item}")
@@ -110,21 +131,30 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             continue
 
         try:
+            if agent_type == "GeminiAgent" and request_count >= MAX_TOTAL_REQUESTS:
+                raise RateLimitError("Global request cap reached")
+
             submitted_answer = agent(question_text)
+            request_count +=1
+
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
 
-            if interrupt_flag:
-                time.sleep(1)
-            else:
-                time.sleep(30) # to not exceed free limits (if still not enough errors, try 60)
+            time.sleep(60 if not interrupt_flag else 1) # to not exceed RPM
 
         except RateLimitError as e:
-            print(f"🛑 TARGET HIT: Gemini Free Tier limit reached.")
+            print(f"🛑 TARGET HIT: Rate limit reached.")
             print(f"Details: {e}")
-            # This is where we break so the space doesn't hang
-            results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": "STOPPED: API LIMIT REACHED"})
-            break
+
+            if agent_type == "GeminiAgent" and active_model_index + 1 < len(GEMINI_MODELS):
+                active_model_index += 1
+                agent = load_gemini_agent(active_model_index)
+                print("▶️ Switched to fallback Gemini model. Retrying question...")
+            else:
+                print("🛑 Stopping and submitting progress.")
+                # This is where we break so the space doesn't hang
+                results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": "STOPPED: API LIMIT REACHED"})
+                break
         
         except Exception as e:
              error_msg = str(e)