app.py

import gradio as gr
import torch
import numpy as np
import json
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from peft import PeftModel
from huggingface_hub import hf_hub_download
import os

# --- 1. CONFIGURATION ---
ADAPTER_REPO = "jvillar-sheff/ag-news-distilbert-lora"
BASE_MODEL_ID = "distilbert-base-uncased"
CLASS_NAMES = {0: "World", 1: "Sports", 2: "Business", 3: "Sci/Tech"}

# --- 2. DYNAMIC METRICS LOADING ---
def fetch_metrics():
    """Downloads evaluation_report.json from the Model Hub."""
    try:
        file_path = hf_hub_download(repo_id=ADAPTER_REPO, filename="evaluation_report.json")
        with open(file_path, "r") as f:
            data = json.load(f)
        
        # Extract numbers
        acc = data['overall_metrics']['Accuracy']
        f1 = data['overall_metrics']['F1 Macro']
        
        return {
            "Accuracy": f"{acc:.2%}",
            "F1_Score": f"{f1:.4f}"
        }
    except Exception as e:
        print(f"Error loading metrics: {e}")
        return {"Accuracy": "N/A", "F1_Score": "N/A"}

# Load metrics on app startup
MODEL_METRICS = fetch_metrics()

# --- 3. MODEL LOADING ---
def load_model():
    print("Loading Base Model...")
    base_model = AutoModelForSequenceClassification.from_pretrained(
        BASE_MODEL_ID,
        num_labels=len(CLASS_NAMES),
        id2label={k: v for k, v in enumerate(CLASS_NAMES.values())},
        label2id={v: k for k, v in CLASS_NAMES.items()}
    )
    
    print("Loading Tokenizer...")
    tokenizer = AutoTokenizer.from_pretrained(ADAPTER_REPO)
    
    print("Loading Adapters...")
    model = PeftModel.from_pretrained(base_model, ADAPTER_REPO)
    
    # Force CPU for Free Tier Spaces
    device = torch.device("cpu")
    model.to(device)
    model.eval()
    return model, tokenizer, device

model, tokenizer, device = load_model()

# --- 4. PREDICTION LOGIC ---
def predict(text):
    if not text.strip():
        return None, None, None

    inputs = tokenizer(
        text, return_tensors="pt", truncation=True, padding="max_length", max_length=128
    ).to(device)

    with torch.no_grad():
        outputs = model(**inputs)

    logits = outputs.logits
    probs = torch.nn.functional.softmax(logits, dim=1).squeeze().cpu().numpy()
    
    # 1. Get Top Label
    pred_idx = np.argmax(probs)
    pred_label = CLASS_NAMES[pred_idx]
    conf = float(probs[pred_idx])

    # 2. Create Probability Dict for the Chart
    class_probs = {CLASS_NAMES[i]: float(probs[i]) for i in range(len(CLASS_NAMES))}

    # 3. Create HTML for the "Confidence Badge"
    if conf > 0.85:
        bg_color, txt_color, icon = "#d4edda", "#155724", "↑" # Green
    elif conf > 0.60:
        bg_color, txt_color, icon = "#fff3cd", "#856404", "~" # Yellow
    else:
        bg_color, txt_color, icon = "#f8d7da", "#721c24", "↓" # Red
        
    badge_html = f"""
    <div style='background-color: {bg_color}; color: {txt_color}; 
    padding: 8px 16px; border-radius: 5px; display: inline-block; font-weight: bold; font-size: 16px;'>
    {icon} Confidence: {conf:.2%}
    </div>
    """
    
    # Return: Label Text, Badge HTML, Chart Data
    return f"# {pred_label}", badge_html, class_probs

# --- 5. UI LAYOUT (gr.Blocks) ---
# Using Soft theme (requires newer Gradio version in requirements.txt)
# If it fails, remove theme=gr.themes.Soft()
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    
    gr.Markdown("# 📰 NLP News Classifier")
    gr.Markdown("Classify news articles into World, Sports, Business, or Sci/Tech using DistilBERT + LoRA.")

    # -- The "Green Banner" (HTML) --
    gr.HTML(f"""
    <div style="background-color: #d1e7dd; color: #0f5132; padding: 15px; border-radius: 5px; border: 1px solid #badbcc; margin-bottom: 20px;">
        ✅ <b>Model Performance (Test Set):</b> Accuracy: {MODEL_METRICS['Accuracy']} | F1 Score: {MODEL_METRICS['F1_Score']}
    </div>
    """)

    with gr.Row():
        # Left Column: Input
        with gr.Column(scale=1):
            input_text = gr.Textbox(
                lines=6, 
                placeholder="Paste a news snippet here...", 
                label="News Article"
            )
            btn = gr.Button("Classify Article", variant="primary")
            
            gr.Markdown("### Examples")
            gr.Examples(
                examples=[
                    ["The stock market rallied today as tech companies reported record profits."],
                    ["The local team won the championship after a stunning overtime goal."],
                    ["NASA announces plans to launch a new rover to Mars next July."]
                ],
                inputs=input_text
            )

        # Right Column: Results
        with gr.Column(scale=1):
            gr.Markdown("### Prediction")
            # Output 1: Big Label text
            out_label = gr.Markdown()
            # Output 2: The Colored Badge
            out_badge = gr.HTML()
            
            gr.Markdown("### Probability Breakdown")
            # Output 3: Bar Chart
            out_chart = gr.Label(num_top_classes=4, label="Confidence Scores")

    # Wire up the button
    btn.click(
        fn=predict,
        inputs=input_text,
        outputs=[out_label, out_badge, out_chart]
    )

# Launch
if __name__ == "__main__":
    demo.launch()