Refactor app.py to integrate AgentWorkflow and update LLM model; add Wikipedia tool to requirements.txt

app.py

@@ -3,31 +3,48 @@ import gradio as gr
 import requests
 import pandas as pd
 
+from llama_index.core.agent.workflow import AgentWorkflow
 from llama_index.llms.huggingface_api import HuggingFaceInferenceAPI
+from llama_index.tools.wikipedia import WikipediaToolSpec
+from llama_index.core.tools import FunctionTool
+from llama_index.tools.duckduckgo import DuckDuckGoSearchToolSpec
+
 
 # (Keep Constants as is)
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 
-llm = HuggingFaceInferenceAPI(model_name='Qwen/Qwen2.5-Coder-32B-Instruct')
+llm = HuggingFaceInferenceAPI(model_name='Qwen/Qwen3-32B', num_output=2048)
+wikipedia_tools = WikipediaToolSpec().to_tool_list()
+search_tool = FunctionTool.from_defaults(DuckDuckGoSearchToolSpec().duckduckgo_full_search)
 
-class TofuAgent:
-    def __init__(self):
-        print("Agent initialized.")
+agent = AgentWorkflow.from_tools_or_functions(
+    tools_or_functions=[search_tool],
+    llm=llm,
+    system_prompt="""You are a concise answer engine.
+ALWAYS output *only* the single entity requested—no explanations, no punctuation beyond the entity itself.
+If unsure, reply “Unknown”.
+Output must be a single word or name or number. Do not output anything else.
+Examples:
+Q: What is the capital of France?
+A: Paris
 
-    def __call__(self, question: str) -> str:
-        print(f"Agent received question (first 50 chars): {question[:50]}...")
-        try:
-            # Use the llm to generate an answer
-            response = llm.complete(question)
-            print(f"Agent generated response: {response}")
-            return response
-        except Exception as e:
-            print(f"Error generating response: {e}")
-            return f"Error: {e}"
+Q: Who nominated the only Featured Article on English Wikipedia about a dinosaur promoted in November 2016?
+A: Ian Rose
+
+Q: How many legs does a spider have?
+A: 8
+
+Do not think step by step. Do not show any internal reasoning. Only respond with the exact entity.
+
+You can use the Wikipedia tool to search for information on Wikipedia, and the DuckDuckGo tool to search the web.
+Some of the questions might be valid sentences but with characters reversed.
+    """,
+    verbose=False,
+)
 
-def run_and_submit_all( profile: gr.OAuthProfile | None):
+async def run_and_submit_all( profile: gr.OAuthProfile | None):
     """
     Fetches all questions, runs the BasicAgent on them, submits all answers,
     and displays the results.
@@ -48,7 +65,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
 
     # 1. Instantiate Agent ( modify this part to create your agent)
     try:
-        agent = TofuAgent()
+        agent = agent
     except Exception as e:
         print(f"Error instantiating agent: {e}")
         return f"Error initializing agent: {e}", None
@@ -88,7 +105,8 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             print(f"Skipping item with missing task_id or question: {item}")
             continue
         try:
-            submitted_answer = agent(question_text)
+            r = await agent.run(question_text)
+            submitted_answer = r.response.blocks[0].text
             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:

requirements.txt

@@ -7,3 +7,4 @@ llama-index-llms-huggingface
 llama-index-llms-huggingface-api
 llama-index-tools-duckduckgo
 llama-index-retrievers-bm25
+llama-index-tools-wikipedia