Upload 3 files

agentic/langgraph_agent.py

@@ -37,6 +37,7 @@ class LangGraphAgent4GAIA:
             subtract,
             divide,
             modulo,
+            add_list,
             web_search,
             arxiv_search,
             wiki_search
@@ -67,7 +68,7 @@ class LangGraphAgent4GAIA:
 
 
         # 2. Bind tools to LLM
-        self.llm_with_tools = llm.bind_tools(tools)
+        self.llm_with_tools = llm.bind_tools(tools, parallel_tool_calls=False)
 
         builder = StateGraph(MessagesState)
         builder.add_node("assistant", self.assistant)
@@ -82,6 +83,7 @@ class LangGraphAgent4GAIA:
         # Compile graph
         return builder.compile()
 
+
 if __name__ == "__main__":
     from langchain_core.runnables.graph import MermaidDrawMethod
 

agentic/tools.py

@@ -2,6 +2,7 @@ from langchain_core.tools import tool
 from langchain_community.tools import DuckDuckGoSearchResults
 from langchain_community.tools.tavily_search import TavilySearchResults
 from langchain_community.document_loaders import ArxivLoader, WikipediaLoader
+from typing import List 
 
 @tool
 def add(a: int, b: int) -> int:
@@ -59,6 +60,13 @@ def modulo(a: int, b: int) -> int:
     """
     return a % b
 
+def add_list(lst: List[float]) -> float:
+    """Sum up a list of numbers.
+
+    Args:
+        lst: A list of numbers.
+    """
+    return sum(lst)
 
 @tool
 def web_search(query: str) -> str:
@@ -66,15 +74,14 @@ def web_search(query: str) -> str:
 
     Args:
         query: The search query."""
-    # search_docs = DuckDuckGoSearchResults(max_results=3).invoke(query=query)
-    search_docs = TavilySearchResults(max_results=3).invoke(query=query)
+    # search_docs = DuckDuckGoSearchResults(max_results=3).invoke({'query': query})
+    search_docs = TavilySearchResults(max_results=3).invoke({'query': query})
     formatted_search_docs = "\n\n---\n\n".join(
         [
             f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content}\n</Document>'
             for doc in search_docs
         ])
-    # return {"web_results": formatted_search_docs}
-    return formatted_search_docs
+    return f"web_results:\n{formatted_search_docs}"
 
 
 @tool
@@ -89,8 +96,7 @@ def arxiv_search(query: str) -> str:
             f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content[:1000]}\n</Document>'
             for doc in search_docs
         ])
-    # return {"arxiv_results": formatted_search_docs}
-    return formatted_search_docs
+    return f"arxiv_content:\n{formatted_search_docs}"
 
 
 @tool
@@ -105,5 +111,4 @@ def wiki_search(query: str) -> str:
             f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content}\n</Document>'
             for doc in search_docs
         ])
-    # return {"wiki_results": formatted_search_docs}
-    return formatted_search_docs
+    return f"wikipedia_content:\n{formatted_search_docs}"

app.py

@@ -16,13 +16,14 @@ hence prompt it well ! The GAIA team shared a prompting example for your agent h
 submission, just make your agent reply with the answer and nothing else).
 """
 import os
+import base64
 import gradio as gr
 import requests
 import pandas as pd
 from omegaconf import OmegaConf
 from langchain_core.messages import HumanMessage
 from agentic import LangGraphAgent4GAIA
-
+from typing import List, Any 
 
 def load_config(config_path: str):
     config = OmegaConf.load(config_path)
@@ -50,14 +51,57 @@ class Agent:
         )
         print("LangGraphAgent4GAIA initialized.")
 
-    def __call__(self, question: str) -> str:
+    @staticmethod
+    def _load_image(file_location: str) -> Any:
+        file_location = file_location.split(".")[0]
+        file_location = f"https://agents-course-unit4-scoring.hf.space/files/{file_location}"
+
+        response = requests.get(file_location)
+        contents = response.content
+        image_base64 = base64.b64encode(contents).decode("utf-8")
+        return image_base64
+
+    @staticmethod
+    def _concat_question_image(question: str, image_base64: Any) -> List[Any]:
+        messages = [
+            HumanMessage(
+                content=[
+                    {
+                        "type": "text",
+                        "text": question
+                    },
+                    {
+                        "type": "image_url",
+                        "image_url": {
+                            "url": f"data:image/png;base64,{image_base64}"
+                        },
+                    },
+                ]
+            )
+        ]
+        return messages
+
+    def _format_message(self, question, file_name):
+        if file_name != "":
+            suffix = file_name.split(".")[-1]
+            if suffix == "png":
+                image_base64 = self._load_image(file_name)
+                messages = self._concat_question_image(question, image_base64)
+            else:
+                question = f"{question} (FilePath: {file_name})"
+                messages = [HumanMessage(content=question)]
+        else:
+            messages = [HumanMessage(content=question)]
+        return messages
+    
+    def __call__(self, question: str, file_name: str) -> str:
         print(f"Agent received question (first 50 chars): {question[:50]}...")
-        messages = [HumanMessage(content=question)]
-        result = self.agent_manager.graph.invoke({"messages": messages}, {"recursion_limit": 50})
+        messages = self._format_message(question, file_name)
+        result = self.agent_manager.graph.invoke({"messages": messages}, {"recursion_limit": 30})
         answer = result['messages'][-1].content
         final_answer = answer.split('FINAL ANSWER: ')[-1].strip()
-        return final_answer
-
+        return result, final_answer
+  
 
 def run_and_submit_all( profile: gr.OAuthProfile | None):
     """
@@ -116,11 +160,14 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
     for item in questions_data:
         task_id = item.get("task_id")
         question_text = item.get("question")
+        file_name = item.get("file_name")
         if not task_id or question_text is None:
             print(f"Skipping item with missing task_id or question: {item}")
             continue
         try:
-            submitted_answer = agent(question_text)
+            results, submitted_answer = agent(question_text, file_name)
+            for msg in results["messages"]:
+                msg.pretty_print()
             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})
         except Exception as e: