Update multiagents.py

multiagents.py

@@ -1,53 +1,73 @@
-# multiagent.py — GAIA-compliant multi-agent system using Groq (patched)
+
+# a multi agent proposal to solve HF agent course final assignment
 import os
 import dotenv
-
-from smolagents import CodeAgent, PythonInterpreterTool
-from smolagents.models.openai_server_model import OpenAIServerModel as BaseOpenAIServerModel
-
+from smolagents import CodeAgent
+from smolagents import OpenAIServerModel
 from tools.fetch import fetch_webpage, search_web
+from smolagents import PythonInterpreterTool
 from tools.yttranscript import get_youtube_transcript, get_youtube_title_description
 from tools.stt import get_text_transcript_from_audio_file
 from tools.image import analyze_image
 from common.mylogger import mylog
 import myprompts
 
-# ✅ Load .env
 dotenv.load_dotenv()
 
-# ✅ Monkeypatch: Ensure message['content'] is always a string
-class PatchedOpenAIServerModel(BaseOpenAIServerModel):
-    def complete_chat(self, messages, **kwargs):
-        for msg in messages:
-            if not isinstance(msg.get("content", ""), str):
-                msg["content"] = str(msg["content"])
-        return super().complete_chat(messages, **kwargs)
-
-# ✅ Groq model (OpenAI-compatible)
-groq_model = PatchedOpenAIServerModel(
-    model_id="llama3-70b-8192",
-    api_key=os.environ["GROQ_API_KEY"],
-    api_base="https://api.groq.com/openai/v1",
+gemini_model = OpenAIServerModel(
+    model_id="gemini-2.0-flash",
+    api_key=os.environ["GEMINI_API_KEY"],
+    # Google Gemini OpenAI-compatible API base URL
+    api_base="https://generativelanguage.googleapis.com/v1beta/openai/",
+)
+
+vllm_model = OpenAIServerModel(
+    model_id="Qwen/Qwen2.5-1.5B-Instruct",
+    api_base="http://192.168.1.39:18000/v1",  
+    api_key="token-abc123",  
+)
+
+openai_41nano_model = OpenAIServerModel(
+    model_id="gpt-4.1-nano",
+    api_base="https://api.openai.com/v1",
+    api_key=os.environ["OPENAI_API_KEY"],
+)
+
+openai_41mini_model = OpenAIServerModel(
+    model_id="gpt-3.5-turbo",
+    api_base="https://api.openai.com/v1",
+    api_key=os.environ["OPENAI_API_KEY"],
 )
 
-# ✅ Final answer checker
-def check_final_answer(final_answer, agent_memory) -> bool:
+
+def check_final_answer(final_answer, agent_memory)  -> bool:
+    """
+    Check if the final answer is correct.
+    basic check on the length of the answer.
+    """
     mylog("check_final_answer", final_answer)
-    return len(str(final_answer)) <= 200
+    # if return answer is more than 200 characters, we will assume it is not correct    
+    if len(str(final_answer)) > 200:
+        return False
+    else:
+        return True
+
 
-# ✅ Sub-agents
 web_agent = CodeAgent(
-    model=groq_model,
-    tools=[search_web, fetch_webpage],
+    model=openai_41nano_model,
+    tools=[
+        search_web,
+        fetch_webpage,                
+    ],
     name="web_agent",
-    description="Uses search engine and scrapes webpages for content.",
-    additional_authorized_imports=["pandas", "numpy", "bs4"],
-    verbosity_level=1,
+    description="Use search engine to find webpages related to a subject and get the page content",
+    additional_authorized_imports=["pandas", "numpy","bs4"],
+    verbosity_level=1,    
     max_steps=7,
 )
 
 audiovideo_agent = CodeAgent(
-    model=groq_model,
+    model=openai_41nano_model,
     tools=[
         get_youtube_transcript,
         get_youtube_title_description,
@@ -55,52 +75,63 @@ audiovideo_agent = CodeAgent(
         analyze_image
     ],
     name="audiovideo_agent",
-    description="Extracts data from audio, video, or images.",
-    additional_authorized_imports=["pandas", "numpy", "bs4", "requests"],
+    description="Extracts information from image, video or audio files from the web",
+    additional_authorized_imports=["pandas", "numpy","bs4", "requests"],
     verbosity_level=1,
     max_steps=7,
 )
 
-# ✅ Manager agent
+
+
 manager_agent = CodeAgent(
-    model=groq_model,
-    tools=[PythonInterpreterTool()],
-    managed_agents=[web_agent, audiovideo_agent],
-    name="manager_agent",
-    description="Coordinates other agents and returns a final answer.",
-    additional_authorized_imports=["pandas", "numpy", "bs4"],
+    model=openai_41mini_model,
+    tools=[ PythonInterpreterTool()],
+    managed_agents=[web_agent, audiovideo_agent],    
+    additional_authorized_imports=["pandas", "numpy","bs4"],
     planning_interval=5,
     verbosity_level=2,
     final_answer_checks=[check_final_answer],
     max_steps=15,
+    name="manager_agent",
+    description="A manager agent that coordinates the work of other agents to answer questions.",
 )
 
-# ✅ Multi-agent interface
 class MultiAgent:
     def __init__(self):
-        print("MultiAgent initialized.")
+        print("BasicAgent initialized.")
 
     def __call__(self, question: str) -> str:
-        mylog(self.__class__.__name__, question)
+        mylog(self.__class__.__name__, question)        
 
         try:
-            prefix = """
-You are the top agent of a multi-agent system that can answer questions by coordinating the work of other agents.
-You can use the web_agent to search the web, or the audiovideo_agent to extract info from audio/video/images.
-You must reason step by step and respect the required output format.
-Only return the final answer in the correct format.
-"""
-            prompt = prefix.strip() + "\nTHE QUESTION:\n" + question.strip() + "\n" + myprompts.output_format.strip()
-            answer = manager_agent.run(prompt)
-            return answer
+            prefix = """You are the top agent of a multi-agent system that can answer questions by coordinating the work of other agents.
+            You will receive a question and you will decide which agent to use to answer it.
+            You can use the web_agent to search the web for information and for fetching the content of a web page, or the audiovideo_agent to extract information from video or audio files.
+            You can also use your own knowledge to answer the question.
+            You need to respect the output format that is given to you.
+            Finding the correct answer to the question need reasoning and plannig, read the question carrefully, think step by step and do not skip any steps.
+            """
+
+            question = prefix + "\nTHE QUESTION:\n" + question + '\n' + myprompts.output_format
+
+            fixed_answer = ""
+
+            fixed_answer = manager_agent.run(question)
+            
+            return fixed_answer
         except Exception as e:
             error = f"An error occurred while processing the question: {e}"
             print(error)
             return error
 
-# ✅ Local test
+
 if __name__ == "__main__":
-    question = "What was the actual enrollment of the Malko competition in 2023?"
+    # Example usage
+   
+    question = """
+What was the actual enrollment of the Malko competition in 2023?
+"""
     agent = MultiAgent()
     answer = agent(question)
     print(f"Answer: {answer}")
+