app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import random

# Configuration 🛠️
model_name = "microsoft/phi-3-mini-4k-instruct"
device = "cuda" if torch.cuda.is_available() else "cpu"

# Load model with optimizations
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16 if device == "cuda" else torch.float32,
    device_map="auto",
    low_cpu_mem_usage=True
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Safety tools 🛡️
SAFE_RESPONSES = [
    "Let's focus on positive tech projects! 🌱",
    "How about designing an eco-friendly robot? 🤖"
]

def generate_response(message, history):
    # Convert history to new message format
    messages = [{"role": "user", "content": msg} for msg, _ in history]
    messages += [{"role": "assistant", "content": res} for _, res in history]
    
    # Safety check
    if any(word in message.lower() for word in ["violence", "hate", "gun"]):
        return random.choice(SAFE_RESPONSES)
    
    # Generate response
    inputs = tokenizer.apply_chat_template(
        [{"role": "user", "content": message}],
        return_tensors="pt"
    ).to(model.device)
    
    outputs = model.generate(
        inputs,
        max_new_tokens=256,
        temperature=0.7,
        do_sample=True
    )
    
    return tokenizer.decode(outputs[0][inputs.shape[-1]:], skip_special_tokens=True)

# Create Gradio interface with updated message format
demo = gr.ChatInterface(
    generate_response,
    examples=[
        "How to make a solar-powered robot?",
        "Python code for air quality sensor"
    ],
    title="🤖 REACT Ethical AI Lab",
    chatbot=gr.Chatbot(height=500)  # Removed invalid 'likeable' parameter
)

if __name__ == "__main__":
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False
    )