Update app.py

app.py

@@ -1,85 +1,82 @@
 import os
-import gradio as gr
 import requests
-import pandas as pd
-
-from smolagents import CodeAgent, DuckDuckGoSearchTool
-from smolagents.models import OpenAIServerModel
-
-from smolagents import Tool
-from wikipedia_searcher import WikipediaSearcher
-
+from smolagents import Agent
 from audio_transcriber import AudioTranscriptionTool
-
 from image_analyzer import ImageAnalysisTool
-
-
-class WikipediaSearchTool(Tool):
-    name = "wikipedia_search"
-    description = "Search Wikipedia for a given query."
-    inputs = {
-        "query": {
-            "type": "string",
-            "description": "The search query string"
-        }
-    }
-    output_type = "string"
-
+from wikipedia_searcher import WikipediaSearcher
+from openai import OpenAI
+from dotenv import load_dotenv
+
+load_dotenv()
+
+# Tools
+audio_tool = AudioTranscriptionTool()
+image_tool = ImageAnalysisTool()
+wiki_tool = WikipediaSearcher()
+
+# Static system prompt
+def build_prompt(question: str) -> str:
+    return f"""You are an agent solving the GAIA benchmark and you are required to provide exact answers.
+Rules to follow:
+1. Return only the exact requested answer: no explanation and no reasoning.
+2. For yes/no questions, return exactly "Yes" or "No".
+3. For dates, use the exact format requested.
+4. For numbers, use the exact number, no other format.
+5. For names, use the exact name as found in sources.
+6. If the question has an associated file, download the file first using the task ID.
+Examples of good responses:
+- "42"
+- "Yes"
+- "October 5, 2001"
+- "Buenos Aires"
+Never include phrases like "the answer is..." or "Based on my research".
+Only return the exact answer.
+
+QUESTION:
+{question}
+"""
+
+# Main agent function
+class GAIAAgent:
     def __init__(self):
-        super().__init__()
-        self.searcher = WikipediaSearcher()
+        self.llm = OpenAI(model="gpt-4-turbo", temperature=0)
 
-    def forward(self, query: str) -> str:
-        return self.searcher.search(query)
-
-wikipedia_search_tool = WikipediaSearchTool()
-
-
-
-# Define the system prompt
-SYSTEM_PROMPT = """You are a general AI assistant. I will ask you a question.
-Report your thoughts, and finish your answer with the following template:
-FINAL ANSWER: [YOUR FINAL ANSWER].
-YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list 
-of numbers and/or strings. If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise. If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string."""
-
-DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
+    def __call__(self, task: dict) -> str:
+        question = task.get("question", "")
+        attachment_url = task.get("attachment", "")
+
+        # Handle audio
+        if attachment_url.endswith((".mp3", ".wav")):
+            transcript = audio_tool.forward(attachment_url)
+            question += f"\n\nTranscript of attached audio:\n{transcript}"
+
+        # Handle image
+        elif attachment_url.endswith((".jpg", ".jpeg", ".png")):
+            return image_tool.forward(attachment_url, question)
+
+        # Handle Python file
+        elif attachment_url.endswith(".py"):
+            try:
+                code_text = requests.get(attachment_url).text
+                question += f"\n\nAttached Python file content:\n{code_text}"
+            except Exception as e:
+                return f"Error retrieving Python file: {e}"
+
+        # Wikipedia queries (if task type or instruction indicates)
+        if "wikipedia" in question.lower():
+            return wiki_tool.search(question)
+
+        # Build prompt
+        prompt = build_prompt(question)
+
+        # Run model
+        response = self.llm.chat.completions.create(
+            messages=[{"role": "system", "content": prompt}],
+            model="gpt-4-turbo"
+        )
 
-# Patched model to prepend system prompt correctly
-class PatchedOpenAIServerModel(OpenAIServerModel):
-    def generate(self, messages, stop_sequences=None, **kwargs):
-        if isinstance(messages, list):
-            if not any(m["role"] == "system" for m in messages):
-                messages = [{"role": "system", "content": SYSTEM_PROMPT}] + messages
-        else:
-            raise TypeError("Expected 'messages' to be a list of message dicts")
-        return super().generate(messages=messages, stop_sequences=stop_sequences, **kwargs)
+        return response.choices[0].message.content.strip()
 
-class MyAgent:
-    def __init__(self):
-        self.model = PatchedOpenAIServerModel(model_id="gpt-4-turbo") #gpt-4-turbo
-        self.agent = CodeAgent(tools=[
-            DuckDuckGoSearchTool(),
-            wikipedia_search_tool,
-            AudioTranscriptionTool(),
-            ImageAnalysisTool()
-        ], model=self.model)
-
-    def __call__(self, task: dict) -> str:
-        question_text = task.get("question", "")
-        
-        # Merge any code or attachment content if available
-        if "code" in task:
-            question_text += f"\n\nAttached code:\n{task['code']}"
-        elif "attachment" in task:
-            question_text += f"\n\nAttached content:\n{task['attachment']}"
-        #Consider audio video
-        #if "L1vXCYZAYYM" in question or "https://www.youtube.com/watch?v=L1vXCYZAYYM" in question:
-            #return "FINAL ANSWER: 11"  # Replace with correct known number
-        if "L1vXCYZAYYM" in question_text or "https://www.youtube.com/watch?v=L1vXCYZAYYM" in question_text:
-            return "FINAL ANSWER: 11"
-
-        return self.agent.run(question_text)
 
 def run_and_submit_all(profile: gr.OAuthProfile | None):
     space_id = os.getenv("SPACE_ID")