VectorV17 commited on
Commit
771f726
·
verified ·
1 Parent(s): 5b1858b

Create app.py

Browse files
Files changed (1) hide show
  1. app.py +102 -0
app.py ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+ import torch
3
+ from transformers import pipeline
4
+ from langchain_community.vectorstores import FAISS
5
+ from langchain_huggingface import HuggingFaceEmbeddings
6
+ from langchain_core.documents import Document
7
+
8
+ # ==========================================
9
+ # 1. INITIALIZE MODELS (CPU OPTIMIZED)
10
+ # ==========================================
11
+ print("Loading Embedding Model...")
12
+ embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
13
+
14
+ print("Loading LLM (TinyLlama) on CPU...")
15
+ # Using TinyLlama as it is small enough to run inference on a basic CPU without crashing
16
+ chat_pipeline = pipeline(
17
+ "text-generation",
18
+ model="TinyLlama/TinyLlama-1.1B-Chat-v1.0",
19
+ device="cpu",
20
+ torch_dtype=torch.float32,
21
+ max_new_tokens=256
22
+ )
23
+
24
+ # ==========================================
25
+ # 2. BUILD FAISS KNOWLEDGE BASE
26
+ # ==========================================
27
+ # In a real app, you would load PDFs or URLs here. We are using mock news data.
28
+ mock_news_data = [
29
+ "AI taking over jobs: Studies show AI will displace 85 million jobs, but create 97 million new ones by 2025.",
30
+ "Remote work: Data indicates remote workers are 13% more productive but struggle with team collaboration.",
31
+ "Electric Vehicles: EVs produce zero tailpipe emissions, but battery mining has significant environmental impacts.",
32
+ "Universal Basic Income: UBI trials show decreased poverty rates but raise concerns about inflation and work disincentives."
33
+ ]
34
+
35
+ print("Building FAISS Index...")
36
+ documents = [Document(page_content=text) for text in mock_news_data]
37
+ vector_store = FAISS.from_documents(documents, embedding_model)
38
+
39
+ # ==========================================
40
+ # 3. DEBATE AGENT LOGIC
41
+ # ==========================================
42
+ def debate_agent(user_message, history):
43
+ # 1. Retrieve relevant facts from FAISS based on the user's stance
44
+ retriever = vector_store.as_retriever(search_kwargs={"k": 1})
45
+ docs = retriever.invoke(user_message)
46
+ context = docs[0].page_content if docs else "No specific news context found."
47
+
48
+ # 2. Construct the Prompt
49
+ # We use TinyLlama's specific chat template format (<|system|>, <|user|>, <|assistant|>)
50
+ system_prompt = (
51
+ "You are an expert Debate Agent. The user will state a topic and their stance. "
52
+ "You must disagree with them, providing a polite counter-argument. "
53
+ "At the very end of your response, score the user's argument strength from 1 to 10 based on logic. "
54
+ f"Use this news fact to support your counter-argument: {context}"
55
+ )
56
+
57
+ prompt = f"<|system|>\n{system_prompt}</s>\n<|user|>\n{user_message}</s>\n<|assistant|>\n"
58
+
59
+ # 3. Generate Response
60
+ try:
61
+ outputs = chat_pipeline(
62
+ prompt,
63
+ do_sample=True,
64
+ temperature=0.7,
65
+ top_k=50,
66
+ top_p=0.95
67
+ )
68
+
69
+ # Extract only the assistant's reply
70
+ generated_text = outputs[0]["generated_text"]
71
+ response = generated_text.split("<|assistant|>\n")[-1].strip()
72
+ return response
73
+
74
+ except Exception as e:
75
+ return f"An error occurred during generation: {str(e)}"
76
+
77
+ # ==========================================
78
+ # 4. GRADIO USER INTERFACE
79
+ # ==========================================
80
+ # Designing a clean UI inspired by the reference images
81
+ with gr.Blocks(theme=gr.themes.Soft()) as demo:
82
+ gr.Markdown(
83
+ """
84
+ # 🤖 AI News Debate & Summary Agent
85
+ **About This Agent:** Enter a topic and your stance (e.g., "I think remote work is bad for productivity").
86
+ The agent will retrieve relevant news facts, formulate a counter-argument, and score the strength of your initial point.
87
+ """
88
+ )
89
+
90
+ chat_interface = gr.ChatInterface(
91
+ fn=debate_agent,
92
+ fill_height=True,
93
+ examples=[
94
+ "I believe AI will eventually destroy all human jobs.",
95
+ "Electric vehicles are the only way to save the environment.",
96
+ "Universal Basic Income will just make people lazy."
97
+ ]
98
+ )
99
+
100
+ if __name__ == "__main__":
101
+ # Launching with parameters suitable for Hugging Face Spaces
102
+ demo.launch(server_name="0.0.0.0", server_port=7860)