app.py

import os
import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from huggingface_hub import snapshot_download, hf_hub_download

# Download model files
def setup_model():
    instruct_repo = "Qwen/Qwen2.5-0.5B-Instruct"
    local_dir = snapshot_download(repo_id=instruct_repo)
    gguf_filename = "qwen2.5-0.5b-instruct-q5_k_m.gguf"
    hf_hub_download(
        repo_id="Qwen/Qwen2.5-0.5B-Instruct-GGUF",
        filename=gguf_filename,
        local_dir=local_dir,
        local_dir_use_symlinks=False
    )
    tokenizer = AutoTokenizer.from_pretrained(local_dir, trust_remote_code=True)
    model = AutoModelForCausalLM.from_pretrained(
        local_dir,
        gguf_file=gguf_filename,
        trust_remote_code=True
    )
    return tokenizer, torch.compile(model)

tokenizer, model = setup_model()


prompt_prefix = """
You are an energy‑saving expert tasked to help households reduce their monthly electricity bills.
Analyze the following appliance usage data, which is provided in the format "Appliance Name: Wattage, hours/day, days/week". 
1. Flag the highest energy consumers.
2. Recommend practical, empathetic, achievable actions.
3. Suggest appliance swaps (e.g. Incandescent to LED, inverter AC) and habit changes.
Provide at most 5 recommendation bullet points and stop there.
Input Data:
"""

def generate_recommendation(appliance_info: str) -> str:
    prompt = prompt_prefix + appliance_info + "\n\nRecommendations:"
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=120,
            use_cache=True,
            do_sample=False,
            temperature=0.0
        )
    text = tokenizer.decode(outputs[0], skip_special_tokens=True)
    # Extract only the part after "Recommendations:"
    recommendation = text.split("Recommendations:")[-1].strip()
    # Keep only the first 5 numbered lines
    lines = recommendation.splitlines()
    filtered_lines = [line for line in lines if line.strip() and line.strip()[0].isdigit()][:5]
    return "\n".join(filtered_lines)


# Define the Gradio interface
iface = gr.Interface(
    fn=generate_recommendation,
    inputs=gr.Textbox(lines=10, placeholder="Enter appliance usage details..."),
    outputs="text",
    title="Energy-Saving Recommendation Generator",
    description="Provide appliance usage details to receive energy-saving tips."
)

if __name__ == "__main__":
    iface.launch()