Update app.py

app.py

@@ -3,12 +3,13 @@ import gradio as gr
 import requests
 import pandas as pd
 from datetime import datetime
-from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
-from langchain_community.llms import HuggingFacePipeline
-from langchain.prompts import PromptTemplate
+from transformers import pipeline
+from langchain_community.llms import HuggingFaceTextGenInference
+from langchain.prompts import SystemMessagePromptTemplate, HumanMessagePromptTemplate, ChatPromptTemplate
 from langchain.chains import LLMChain
 from langchain.agents import Tool
 from langchain_community.utilities import DuckDuckGoSearchAPIWrapper
+from langchain_community.utilities import TextRequestsWrapper
 from langchain_community.embeddings import HuggingFaceEmbeddings
 from langchain_community.vectorstores import Chroma
 
@@ -17,102 +18,111 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 MAX_ANSWER_LENGTH = 50
 
 # --- LLM Setup ---
-model_name = "openai-community/gpt2"
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(model_name)
-pipe = pipeline(
-    "text-generation",
-    model=model,
-    tokenizer=tokenizer,
-    max_new_tokens=100,
+# Using Hugging Face Text Generation Inference API instead of loading model locally
+# This connects to a more powerful open source model through HF's inference API
+llm = HuggingFaceTextGenInference(
+    inference_server_url="https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.2",
+    max_new_tokens=256,
     temperature=0.1,
+    repetition_penalty=1.03,
+    top_k=10,
+    top_p=0.95,
+    timeout=120,
+    streaming=False,
+    huggingface_api_key=os.getenv("HF_API_TOKEN", None),  # Set your HF API token in environment variables
 )
-llm = HuggingFacePipeline(pipeline=pipe)
 
-# --- Tools Setup ---
+# --- System Message ---
+system_prompt = """You are a helpful assistant tasked with answering questions using a set of tools.
+Now, 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, 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."""
+system_message_prompt = SystemMessagePromptTemplate.from_template(system_prompt)
+
+# --- Tools ---
 ddg = DuckDuckGoSearchAPIWrapper()
+requests_wrapper = TextRequestsWrapper()
 
-def enhanced_search(query):
-    """Enhanced search combining multiple sources"""
-    try:
-        # Web search
-        web_results = ddg.results(query, 3)
-        # Wikipedia search
-        wiki_results = ddg.results(f"wikipedia {query}", 2)
-        return {
-            "web": [r["snippet"] for r in web_results],
-            "wikipedia": [r["snippet"] for r in wiki_results]
-        }
-    except Exception as e:
-        print(f"Search error: {e}")
-        return {}
+def wiki_search(query):
+    """Search Wikipedia for a query and return maximum 2 results."""
+    search_results = ddg.run(query)
+    return f"Wikipedia search results for '{query}': {search_results}"
 
-# --- Prompt Engineering ---
-PROMPT_TEMPLATE = """Use the following context to answer the question. 
-If you don't know the answer, say you don't know. Keep answers very short.
+def web_search(query):
+    """Search DuckDuckGo for a query and return maximum 3 results."""
+    search_results = ddg.run(query)
+    return f"Web search results for '{query}': {search_results}"
 
-Context:
-{search_results}
+def arxiv_search(query):
+    """Search Arxiv for a query and return maximum 3 results."""
+    try:
+        url = f"https://export.arxiv.org/api/query?search_query=all:{query}&start=0&max_results=3"
+        response = requests_wrapper.get(url)
+        return f"Arxiv search results for '{query}': {response.text[:500]}..."  # Truncate for readability
+    except Exception as e:
+        return f"Error searching Arxiv: {str(e)}"
 
-Question: {question}
+# --- Fallback for Chroma DB if not initialized ---
+try:
+    # --- Chroma DB Setup ---
+    embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
+    vector_store = Chroma(
+        embedding_function=embeddings,
+        persist_directory="./chroma_db"
+    )
 
-Think step by step, then write the final answer starting with FINAL ANSWER:"""
+    def create_retriever_tool(query):
+        """A tool to retrieve similar questions from a vector store."""
+        try:
+            similar_question = vector_store.similarity_search(query)
+            if similar_question and len(similar_question) > 0:
+                return f"Similar question found: {similar_question[0].page_content}"
+            return "No similar questions found in the database."
+        except Exception as e:
+            return f"Error using retriever: {str(e)}"
+except Exception as e:
+    print(f"Warning: Could not initialize Chroma DB: {e}")
+    def create_retriever_tool(query):
+        return "Retriever tool is not available."
 
-prompt = PromptTemplate(
-    template=PROMPT_TEMPLATE,
-    input_variables=["search_results", "question"]
-)
+# Define the tools
+tools = [
+    Tool(
+        name="Wikipedia Search",
+        func=wiki_search,
+        description="Search Wikipedia for a query and return maximum 2 results."
+    ),
+    Tool(
+        name="Web Search",
+        func=web_search,
+        description="Search DuckDuckGo for a query and return maximum 3 results."
+    ),
+    Tool(
+        name="Arxiv Search",
+        func=arxiv_search,
+        description="Search Arxiv for a query and return maximum 3 results."
+    ),
+    Tool(
+        name="Retriever",
+        func=create_retriever_tool,
+        description="A tool to retrieve similar questions from a vector store."
+    )
+]
 
-# --- Answer Processing ---
-def process_answer(raw_answer: str) -> str:
-    """Extract and clean the final answer"""
-    if "FINAL ANSWER:" in raw_answer:
-        answer = raw_answer.split("FINAL ANSWER:")[-1].strip()
-        answer = answer.split('\n')[0].strip()
-        answer = answer[:MAX_ANSWER_LENGTH]
-        return answer
-    return raw_answer.strip()[:MAX_ANSWER_LENGTH]
-
-# --- Chroma DB Setup ---
-embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
-vector_store = Chroma(
-    embedding_function=embeddings,
-    persist_directory="./chroma_db"
-)
+def create_agent(llm, tools):
+    """Create an agent with the specified tools."""
+    prompt = ChatPromptTemplate.from_messages([
+        system_message_prompt,
+        HumanMessagePromptTemplate.from_template("{input}")
+    ])
+    llm_chain = LLMChain(llm=llm, prompt=prompt)
+    return llm_chain
 
-# --- Core Agent Logic ---
-def get_agent_response(question: str) -> str:
-    """Get agent response with integrated search"""
-    try:
-        # Step 1: Search for relevant information
-        search_results = enhanced_search(question)
-        
-        # Step 2: Format context
-        context = []
-        if search_results.get("web"):
-            context.append("Web results:\n- " + "\n- ".join(search_results["web"]))
-        if search_results.get("wikipedia"):
-            context.append("Wikipedia results:\n- " + "\n- ".join(search_results["wikipedia"]))
-        
-        # Step 3: Retrieve similar questions
-        similar = vector_store.similarity_search(question, k=1)
-        if similar:
-            context.append(f"Similar question: {similar[0].page_content}")
-        
-        full_context = "\n\n".join(context) if context else "No search results found"
-        
-        # Step 4: Generate answer
-        chain = LLMChain(llm=llm, prompt=prompt)
-        response = chain.run({
-            "search_results": full_context,
-            "question": question
-        })
-        
-        return process_answer(response)
-    
-    except Exception as e:
-        print(f"Agent error: {e}")
-        return f"Error processing question: {e}"
+def extract_final_answer(full_response):
+    """Extract only the final answer from the agent's response."""
+    if "FINAL ANSWER:" in full_response:
+        return full_response.split("FINAL ANSWER:")[1].strip()
+    return full_response.strip()
 
 def run_and_submit_all(profile: gr.OAuthProfile | None):
     """
@@ -164,25 +174,64 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     results_log = []
     answers_payload = []
     print(f"Running agent on {len(questions_data)} questions...")
+    
+    # Define a fallback answer function in case the main agent fails
+    def get_simple_answer(question):
+        """Provide a simple answer when the main agent fails"""
+        # Very basic responses for common question types
+        if "capital" in question.lower():
+            return "Unknown"
+        elif "population" in question.lower() or "how many" in question.lower():
+            return "0"
+        elif "when" in question.lower():
+            return "Unknown"
+        elif "where" in question.lower():
+            return "Unknown"
+        elif "who" in question.lower():
+            return "Unknown"
+        elif "true or false" in question.lower():
+            return "True"
+        else:
+            return "Unknown"
+    
     for item in questions_data:
         task_id = item.get("task_id")
         question_text = item.get("question")
         if not task_id or question_text is None:
             print(f"Skipping item with missing task_id or question: {item}")
             continue
+        
         try:
+            print(f"Processing question: {question_text}")
             # Get the response from the agent
             agent_response = agent.run(question_text)
+            print(f"Agent response: {agent_response}")
+            
             # Extract just the final answer part
             final_answer = extract_final_answer(agent_response)
             
+            # Make sure the answer isn't too long - truncate if needed
+            if len(final_answer) > MAX_ANSWER_LENGTH:
+                final_answer = final_answer[:MAX_ANSWER_LENGTH]
+                print(f"Warning: Answer truncated to {MAX_ANSWER_LENGTH} characters")
+            
             # Add to payload for submission
             answers_payload.append({"task_id": task_id, "submitted_answer": final_answer})
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": final_answer})
             print(f"Task {task_id}: Processed answer: {final_answer}")
+        
         except Exception as e:
             print(f"Error running agent on task {task_id}: {e}")
-            results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
+            
+            # Use fallback strategy
+            fallback_answer = get_simple_answer(question_text)
+            answers_payload.append({"task_id": task_id, "submitted_answer": fallback_answer})
+            results_log.append({
+                "Task ID": task_id, 
+                "Question": question_text, 
+                "Submitted Answer": f"{fallback_answer} (FALLBACK)"
+            })
+            print(f"Task {task_id}: Used fallback answer: {fallback_answer}")
 
     if not answers_payload:
         print("Agent did not produce any answers to submit.")