create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
from langchain_huggingface import HuggingFacePipeline
|
| 3 |
+
from langchain.agents import initialize_agent, Tool
|
| 4 |
+
from langchain_community.tools import DuckDuckGoSearchRun
|
| 5 |
+
from langchain.memory import ConversationBufferMemory
|
| 6 |
+
from langchain.agents import AgentType
|
| 7 |
+
|
| 8 |
+
# Load the free open-source LLM (Phi-2 runs on CPU, might be a bit slow but works for free hosting)
|
| 9 |
+
llm = HuggingFacePipeline.from_model_id(
|
| 10 |
+
model_id="microsoft/phi-2",
|
| 11 |
+
task="text-generation",
|
| 12 |
+
pipeline_kwargs={
|
| 13 |
+
"max_new_tokens": 256, # Limit response length
|
| 14 |
+
"do_sample": True,
|
| 15 |
+
"temperature": 0.7, # Creativity level
|
| 16 |
+
"top_k": 50,
|
| 17 |
+
"top_p": 0.95
|
| 18 |
+
},
|
| 19 |
+
model_kwargs={"trust_remote_code": True} # Required for Phi-2
|
| 20 |
+
)
|
| 21 |
+
|
| 22 |
+
# Define tools the agent can use (e.g., web search for current info)
|
| 23 |
+
search = DuckDuckGoSearchRun()
|
| 24 |
+
tools = [
|
| 25 |
+
Tool(
|
| 26 |
+
name="Web Search",
|
| 27 |
+
func=search.run,
|
| 28 |
+
description="Useful for answering questions about current events, facts, or anything requiring up-to-date web information. Input should be a search query."
|
| 29 |
+
)
|
| 30 |
+
]
|
| 31 |
+
|
| 32 |
+
# Set up memory to remember conversation history
|
| 33 |
+
memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
|
| 34 |
+
|
| 35 |
+
# Initialize the conversational agent
|
| 36 |
+
agent = initialize_agent(
|
| 37 |
+
tools,
|
| 38 |
+
llm,
|
| 39 |
+
agent=AgentType.CONVERSATIONAL_REACT_DESCRIPTION, # Agent type that handles conversations and tools
|
| 40 |
+
verbose=True, # Logs reasoning (visible in console, not user-facing)
|
| 41 |
+
memory=memory # Enables context understanding across messages
|
| 42 |
+
)
|
| 43 |
+
|
| 44 |
+
# Gradio chat function: Handles user input and gets agent response
|
| 45 |
+
def chat_with_agent(message, history):
|
| 46 |
+
try:
|
| 47 |
+
response = agent.invoke({"input": message})["output"]
|
| 48 |
+
except Exception as e:
|
| 49 |
+
response = f"Error: {str(e)}. Try rephrasing your question."
|
| 50 |
+
return response
|
| 51 |
+
|
| 52 |
+
# Create the Gradio chat interface
|
| 53 |
+
iface = gr.ChatInterface(
|
| 54 |
+
fn=chat_with_agent,
|
| 55 |
+
title="Free Cloud AI Agent",
|
| 56 |
+
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.",
|
| 57 |
+
examples=["What's the latest news on AI?", "Tell me a joke.", "Remember my name is Alex. What's my name?"] # Demo prompts
|
| 58 |
+
)
|
| 59 |
+
|
| 60 |
+
# Launch the app (Hugging Face handles this automatically)
|
| 61 |
+
if __name__ == "__main__":
|
| 62 |
+
iface.launch()
|