Update app.py

app.py

@@ -1,62 +1,65 @@
 import gradio as gr
 from langchain_huggingface import HuggingFacePipeline
-from langchain.agents import initialize_agent, Tool
-from langchain_community.tools import DuckDuckGoSearchRun
+from langchain.chains import ConversationChain
 from langchain.memory import ConversationBufferMemory
-from langchain.agents import AgentType
+from langchain_community.tools import DuckDuckGoSearchRun
+from langchain_core.tools import Tool
 
-# Load the free open-source LLM (Phi-2 runs on CPU, might be a bit slow but works for free hosting)
+# Load a smaller, faster model (TinyLlama-1.1B-Chat, optimized for CPU)
 llm = HuggingFacePipeline.from_model_id(
-    model_id="microsoft/phi-2",
+    model_id="TinyLlama/TinyLlama-1.1B-Chat-v1.0",
     task="text-generation",
     pipeline_kwargs={
-        "max_new_tokens": 256,  # Limit response length
+        "max_new_tokens": 150,  # Shorter responses for speed
         "do_sample": True,
-        "temperature": 0.7,    # Creativity level
-        "top_k": 50,
-        "top_p": 0.95
+        "temperature": 0.6,    # Slightly less creative for consistency
+        "top_k": 40,
+        "top_p": 0.9
     },
-    model_kwargs={"trust_remote_code": True}  # Required for Phi-2
+    model_kwargs={"trust_remote_code": True}
 )
 
-# Define tools the agent can use (e.g., web search for current info)
+# Set up conversation memory
+memory = ConversationBufferMemory()
+
+# Optional: Web search tool (comment out if you don’t need it for faster responses)
 search = DuckDuckGoSearchRun()
 tools = [
     Tool(
         name="Web Search",
         func=search.run,
-        description="Useful for answering questions about current events, facts, or anything requiring up-to-date web information. Input should be a search query."
+        description="Use for current events or facts. Input a search query."
     )
 ]
 
-# Set up memory to remember conversation history
-memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
-
-# Initialize the conversational agent
-agent = initialize_agent(
-    tools,
-    llm,
-    agent=AgentType.CONVERSATIONAL_REACT_DESCRIPTION,  # Agent type that handles conversations and tools
-    verbose=True,  # Logs reasoning (visible in console, not user-facing)
-    memory=memory  # Enables context understanding across messages
+# Create a simple conversation chain (faster than full agent)
+conversation = ConversationChain(
+    llm=llm,
+    memory=memory,
+    verbose=False  # Disable logging for speed
 )
 
-# Gradio chat function: Handles user input and gets agent response
+# Function to handle chat and optional tool use
 def chat_with_agent(message, history):
     try:
-        response = agent.invoke({"input": message})["output"]
+        # Check if the query likely needs a web search (e.g., current events)
+        if any(keyword in message.lower() for keyword in ["latest", "news", "current", "today", "weather"]):
+            search_result = search.run(message)
+            prompt = f"User asked: {message}\nWeb search result: {search_result}\nAnswer based on this info."
+            response = conversation.predict(input=prompt)
+        else:
+            response = conversation.predict(input=message)
     except Exception as e:
         response = f"Error: {str(e)}. Try rephrasing your question."
     return response
 
-# Create the Gradio chat interface
+# Gradio chat interface
 iface = gr.ChatInterface(
     fn=chat_with_agent,
-    title="Free Cloud AI Agent",
-    description="A conversational AI agent that remembers our talks and can search the web for info. Powered by Phi-2, LangChain, and hosted free on Hugging Face Spaces. Responses may take 10-30 seconds on free CPU.",
-    examples=["What's the latest news on AI?", "Tell me a joke.", "Remember my name is Alex. What's my name?"]  # Demo prompts
+    title="Fast Free AI Agent",
+    description="A lightweight conversational AI that remembers our talks and can search the web. Hosted free on Hugging Face Spaces. Responses should be faster (5-15 seconds).",
+    examples=["What's my name if I told you it's Alex?", "Tell me a quick joke.", "What's the latest AI news?"]
 )
 
-# Launch the app (Hugging Face handles this automatically)
 if __name__ == "__main__":
     iface.launch()