app.py

import os
import gradio as gr
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

# Model initialization

def setup_model():
    # Load base model in 4-bit or full precision depending on availability
    base = AutoModelForCausalLM.from_pretrained(
        "Qwen/Qwen-2.5B-0.5", device_map="auto", trust_remote_code=True
    )
    # Load LoRA adapters
    model = PeftModel.from_pretrained(
        base,
        "Wh1plashR/qwen-energy-lora",
        device_map="auto",
        trust_remote_code=True
    )
    # Load tokenizer
    tokenizer = AutoTokenizer.from_pretrained("Wh1plashR/qwen-energy-lora", use_fast=True)

    # Set to eval and optionally compile (requires PyTorch 2+)
    model.eval()
    try:
        model = torch.compile(model)
    except Exception:
        pass

    return tokenizer, model

# Initialize
tokenizer, model = setup_model()

# Prompt prefix
prompt_prefix = """
You are an energy‑saving expert tasked to help households reduce their monthly electricity bills.
Given the user's appliance usage information (device name, wattage, hours used per day, days used per week):
1. Flag the highest energy consumers.
2. Recommend practical, empathetic, achievable actions.
3. Suggest appliance swaps (e.g. LED, inverter AC) and habit changes.
Give at most 5 suggestions and format with bullet points that is <= 100 tokens.
Don't add anything to the response besides the recommendation.
Here is the user's input:
"""

def generate_recommendation(appliance_info: str) -> str:
    # Build prompt
    prompt = prompt_prefix + appliance_info.strip() + "\n\nRecommendations:"
    # Tokenize and move to model device
    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, padding=True).to(model.device)

    # Generate
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=100,
            do_sample=False,
            temperature=0.0,
            use_cache=True
        )

    # Decode and clean
    text = tokenizer.decode(outputs[0], skip_special_tokens=True)
    # Extract recommendations
    rec = text.split("Recommendations:")[-1].strip()
    # Remove any trailing notes
    rec = rec.split("Note:")[0].strip()
    # Clean empty lines
    cleaned = "\n".join(line.strip() for line in rec.splitlines() if line.strip())
    return cleaned

# Gradio interface
def main():
    iface = gr.Interface(
        fn=generate_recommendation,
        inputs=gr.Textbox(lines=10, placeholder="Enter appliance usage details..."),
        outputs=gr.Textbox(label="Energy-Saving Recommendations"),
        title="Energy-Saving Recommendation Generator",
        description="Provide appliance usage details to receive actionable energy-saving tips."
    )
    iface.launch()

if __name__ == "__main__":
    main()