Update agent.py

agent.py

@@ -18,8 +18,12 @@ from smolagents import InferenceClientModel, LiteLLMModel, ToolCallingAgent, Too
 load_dotenv()
 configure(api_key=os.getenv("GOOGLE_API_KEY"))
 
+# Logging
+#logging.basicConfig(level=logging.INFO, format="%(asctime)s | %(levelname)s | %(message)s")
+#logger = logging.getLogger(__name__)
+
 # --- Model Configuration ---
-GEMINI_MODEL_NAME = "gemini/gemini-1.5-flash"
+GEMINI_MODEL_NAME = "gemini/gemini-2.0-flash"
 OPENAI_MODEL_NAME = "openai/gpt-4o"
 GROQ_MODEL_NAME = "groq/llama3-70b-8192"
 DEEPSEEK_MODEL_NAME = "deepseek/deepseek-chat"
@@ -142,10 +146,6 @@ class FileAttachmentQueryTool(Tool):
     }
     output_type = "string"
 
-    def __init__(self, model_name, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.model_name = model_name
-
     def forward(self, task_id: str, mime_type: str | None, user_query: str) -> str:
         file_url = f"https://agents-course-unit4-scoring.hf.space/files/{task_id}"
         file_response = requests.get(file_url)
@@ -163,36 +163,12 @@ class FileAttachmentQueryTool(Tool):
 
         return response.text
 
-class CheatTool(Tool):
-    name = "cheat_tool"
-    description = "Search the GAIA QA dataset for a known answer if question is similar."
-    inputs = {"query": {"type": "string", "description": "The user question."}}
-    output_type = "string"
-
-    def forward(self, query: str) -> str:
-        from difflib import SequenceMatcher
-        try:
-            df = pd.read_csv("gaia_qa.csv")
-            best_match = ""
-            best_score = 0.0
-            for _, row in df.iterrows():
-                score = SequenceMatcher(None, query.strip().lower(), str(row["question"]).strip().lower()).ratio()
-                if score > best_score:
-                    best_score = score
-                    best_match = str(row["answer"]).strip()
-            if best_score > 0.5:
-                return best_match
-            return ""
-        except Exception as e:
-            return f"CheatTool error: {e}"
-
 # --- Basic Agent Definition ---
 class BasicAgent:
     def __init__(self, provider="deepseek"):
         print("BasicAgent initialized.")
         model = self.select_model(provider)
         client = InferenceClientModel()
-        self.cheat_tool = CheatTool()
         tools = [
             DuckDuckGoSearchTool(),
             GeminiVideoQA(GEMINI_MODEL_NAME),
@@ -207,11 +183,11 @@ class BasicAgent:
             model=model,
             tools=tools,
             add_base_tools=False,
-            max_steps=5,
+            max_steps=12,
         )
         self.agent.system_prompt = (
             """
-            You are a GAIA benchmark AI assistant. Your sole purpose is to provide exact, minimal answers in the format 'FINAL ANSWER: [ANSWER]' with no additional text, explanations, or comments. 
+            You are a GAIA benchmark AI assistant. Your sole purpose is to provide exact, minimal answers in the format '[ANSWER]' with no additional text, explanations, or comments. 
 
             - If the answer is a number, use numerals (e.g., '42', not 'forty-two'), without commas or units (e.g., no '$', '%') unless explicitly requested.
             - If the answer is a string, use no articles ('a', 'the'), no abbreviations (e.g., 'New York', not 'NY'), and write digits as text (e.g., 'one', not '1') unless specified.
@@ -221,14 +197,14 @@ class BasicAgent:
             - For Wikipedia or search tools, distill results to the minimal correct answer, ignoring extraneous content.
             - If proving something, compute step-by-step internally but output only the final result in the required format.
             - If tool outputs are verbose, extract only the essential answer that satisfies the question.
-            - Under no circumstances include explanations, intermediate steps, or text outside the 'FINAL ANSWER: [ANSWER]' format.
+            - Under no circumstances include explanations, intermediate steps, or text outside the '[ANSWER]' format.
 
             Example:
             Question: What is 2 + 2?
-            Response: FINAL ANSWER: 4
+            Response: 4
 
             Your response must always be:
-            FINAL ANSWER: [ANSWER]
+            [ANSWER]
             """
         )
 
@@ -246,18 +222,10 @@ class BasicAgent:
 
     def __call__(self, question: str) -> str:
         print(f"Agent received question (first 50 chars): {question[:50]}...")
-
-        cheat_result = self.cheat_tool.forward(question)
-        if cheat_result:
-            return f"FINAL ANSWER: {cheat_result}"
-
         result = self.agent.run(question)
-        if isinstance(result, dict) and "final_answer" in result and isinstance(result["final_answer"], str):
-            final_str = result["final_answer"].strip()
-        else:
-            final_str = str(result).strip()
+        final_str = str(result).strip()
 
-        return f"FINAL ANSWER: {final_str}"
+        return final_str
 
     def evaluate_random_questions(self, csv_path: str = "gaia_qa.csv", sample_size: int = 3, show_steps: bool = True):
         df = pd.read_csv(csv_path)
@@ -277,7 +245,7 @@ class BasicAgent:
                 print("Agent:", result)
                 print("Correct:", expected == result)
             else:
-                print(f"Q: {question}\nE: {expected}\nA: {result}\n\u2713: {expected == result}\n")
+                print(f"Q: {question}\nE: {expected}\nA: {result}\n✓: {expected == result}\n")
 
 if __name__ == "__main__":
     args = sys.argv[1:]