app.py

import streamlit as st
import torch
import torch.hub
import re
import os
import time

# --- Set Page Config First ---
st.set_page_config(
    page_title="AI Text Detector",
    layout="centered",
    initial_sidebar_state="collapsed"
)

# --- Improved CSS for a cleaner UI ---
st.markdown("""
<style>
    /* Modern clean font for the entire app */
    @import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&display=swap');
    
    html, body, [class*="css"] {
        font-family: 'Inter', sans-serif;
    }
    
    /* Header styling */
    h1 {
        font-weight: 700;
        color: #1E3A8A;
        padding-bottom: 1rem;
        border-bottom: 2px solid #E5E7EB;
        margin-bottom: 2rem;
    }
    
    /* Text area styling */
    .stTextArea textarea {
        border: 1px solid #D1D5DB;
        border-radius: 8px;
        font-size: 16px;
        padding: 12px;
        background-color: #F9FAFB;
        box-shadow: 0 1px 2px rgba(0, 0, 0, 0.05);
        transition: border-color 0.15s ease-in-out, box-shadow 0.15s ease-in-out;
    }
    
    .stTextArea textarea:focus {
        border-color: #3B82F6;
        box-shadow: 0 0 0 3px rgba(59, 130, 246, 0.3);
        outline: none;
    }
    
    /* Button styling */
    .stButton button {
        border-radius: 8px;
        font-weight: 600;
        padding: 10px 16px;
        background-color: #2563EB;
        color: white;
        border: none;
        width: 100%;
        transition: background-color 0.2s ease;
    }
    
    .stButton button:hover {
        background-color: #1D4ED8;
    }
    
    /* Result box styling */
    .result-box {
        border-radius: 8px;
        padding: 20px;
        margin-top: 24px;
        text-align: center;
        background-color: white;
        box-shadow: 0 1px 3px rgba(0, 0, 0, 0.1), 0 1px 2px rgba(0, 0, 0, 0.06);
        border: 1px solid #E5E7EB;
    }
    
    /* Result highlights */
    .highlight-human {
        color: #059669;
        font-weight: 600;
        background: rgba(5, 150, 105, 0.1);
        padding: 4px 10px;
        border-radius: 8px;
        display: inline-block;
    }
    
    .highlight-ai {
        color: #DC2626;
        font-weight: 600;
        background: rgba(220, 38, 38, 0.1);
        padding: 4px 10px;
        border-radius: 8px;
        display: inline-block;
    }
    
    /* Footer styling */
    .footer {
        text-align: center;
        margin-top: 40px;
        padding-top: 20px;
        border-top: 1px solid #E5E7EB;
        color: #6B7280;
        font-size: 14px;
    }
    
    /* Progress bar styling */
    .stProgress > div > div {
        background-color: #2563EB;
    }
    
    /* General spacing */
    .block-container {
        padding-top: 2rem;
        padding-bottom: 2rem;
    }
</style>
""", unsafe_allow_html=True)

# --- Configuration ---
MODEL1_PATH = "modernbert.bin"
MODEL2_URL = "https://huggingface.co/mihalykiss/modernbert_2/resolve/main/Model_groups_3class_seed12"
MODEL3_URL = "https://huggingface.co/mihalykiss/modernbert_2/resolve/main/Model_groups_3class_seed22"
BASE_MODEL = "answerdotai/ModernBERT-base"
NUM_LABELS = 41
HUMAN_LABEL_INDEX = 24
DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# --- Model Loading Functions ---
@st.cache_resource(show_spinner=False)
def load_tokenizer(model_name):
    from transformers import AutoTokenizer
    return AutoTokenizer.from_pretrained(model_name)

@st.cache_resource(show_spinner=False)
def load_model(model_path_or_url, base_model, num_labels, is_url=False, _device=DEVICE):
    from transformers import AutoModelForSequenceClassification
    
    # Load base model architecture
    model = AutoModelForSequenceClassification.from_pretrained(base_model, num_labels=num_labels)
    
    try:
        # Load weights
        if is_url:
            state_dict = torch.hub.load_state_dict_from_url(model_path_or_url, map_location=_device, progress=False)
        else:
            if not os.path.exists(model_path_or_url):
                return None
            state_dict = torch.load(model_path_or_url, map_location=_device, weights_only=False)
            
        model.load_state_dict(state_dict)
        model.to(_device).eval()
        return model
    except Exception:
        return None

# --- Text Processing Functions ---
def clean_text(text):
    if not isinstance(text, str):
        return ""
    text = text.replace("\r\n", "\n").replace("\r", "\n")
    text = re.sub(r"\n\s*\n+", "\n\n", text)
    text = re.sub(r"[ \t]+", " ", text)
    text = re.sub(r"(\w+)-\s*\n\s*(\w+)", r"\1\2", text)
    text = re.sub(r"(?<!\n)\n(?!\n)", " ", text)
    return text.strip()

def classify_text(text, tokenizer, model_1, model_2, model_3, device, label_mapping, human_label_index):
    if not all([model_1, model_2, model_3, tokenizer]):
        return {"error": True, "message": "Models failed to load properly."}

    cleaned_text = clean_text(text)
    if not cleaned_text:
        return None

    try:
        inputs = tokenizer(
            cleaned_text,
            return_tensors="pt",
            truncation=True,
            padding=True,
            max_length=tokenizer.model_max_length
        ).to(device)

        with torch.no_grad():
            logits_1 = model_1(**inputs).logits
            logits_2 = model_2(**inputs).logits
            logits_3 = model_3(**inputs).logits

            softmax_1 = torch.softmax(logits_1, dim=1)
            softmax_2 = torch.softmax(logits_2, dim=1)
            softmax_3 = torch.softmax(logits_3, dim=1)

            averaged_probabilities = (softmax_1 + softmax_2 + softmax_3) / 3
            probabilities = averaged_probabilities[0].cpu()

            if not (0 <= human_label_index < len(probabilities)):
                return {"error": True, "message": "Configuration error."}

            human_prob = probabilities[human_label_index].item() * 100

            mask = torch.ones_like(probabilities, dtype=torch.bool)
            mask[human_label_index] = False
            ai_total_prob = probabilities[mask].sum().item() * 100

            ai_probs_only = probabilities.clone()
            ai_probs_only[human_label_index] = -float('inf')
            ai_argmax_index = torch.argmax(ai_probs_only).item()
            ai_argmax_model = label_mapping.get(ai_argmax_index, f"Unknown AI (Index {ai_argmax_index})")

            if human_prob >= ai_total_prob:
                return {"is_human": True, "probability": human_prob, "model": "Human"}
            else:
                return {"is_human": False, "probability": ai_total_prob, "model": ai_argmax_model}

    except Exception as e:
        return {"error": True, "message": f"Analysis failed: {str(e)}"}

# --- Label Mapping ---
LABEL_MAPPING = {
    0: '13B', 1: '30B', 2: '65B', 3: '7B', 4: 'GLM130B', 5: 'bloom_7b',
    6: 'bloomz', 7: 'cohere', 8: 'davinci', 9: 'dolly', 10: 'dolly-v2-12b',
    11: 'flan_t5_base', 12: 'flan_t5_large', 13: 'flan_t5_small',
    14: 'flan_t5_xl', 15: 'flan_t5_xxl', 16: 'gemma-7b-it', 17: 'gemma2-9b-it',
    18: 'gpt-3.5-turbo', 19: 'gpt-35', 20: 'gpt4', 21: 'gpt4o',
    22: 'gpt_j', 23: 'gpt_neox', 24: 'human', 25: 'llama3-70b', 26: 'llama3-8b',
    27: 'mixtral-8x7b', 28: 'opt_1.3b', 29: 'opt_125m', 30: 'opt_13b',
    31: 'opt_2.7b', 32: 'opt_30b', 33: 'opt_350m', 34: 'opt_6.7b',
    35: 'opt_iml_30b', 36: 'opt_iml_max_1.3b', 37: 't0_11b', 38: 't0_3b',
    39: 'text-davinci-002', 40: 'text-davinci-003'
}

# --- Main UI ---
st.title("AI Text Detector")

# Initialization with a progress bar
with st.spinner(""):
    # Create a progress bar
    progress_bar = st.progress(0)
    st.info("Initializing AI detection models...")
    
    # Step 1: Load tokenizer
    progress_bar.progress(20)
    time.sleep(0.5)  # Small delay for visual feedback
    TOKENIZER = load_tokenizer(BASE_MODEL)
    
    # Step 2: Load first model
    progress_bar.progress(40)
    time.sleep(0.5)  # Small delay for visual feedback
    MODEL_1 = load_model(MODEL1_PATH, BASE_MODEL, NUM_LABELS, is_url=False, _device=DEVICE)
    
    # Step 3: Load second model
    progress_bar.progress(60)
    time.sleep(0.5)  # Small delay for visual feedback
    MODEL_2 = load_model(MODEL2_URL, BASE_MODEL, NUM_LABELS, is_url=True, _device=DEVICE)
    
    # Step 4: Load third model
    progress_bar.progress(80)
    time.sleep(0.5)  # Small delay for visual feedback
    MODEL_3 = load_model(MODEL3_URL, BASE_MODEL, NUM_LABELS, is_url=True, _device=DEVICE)
    
    # Complete initialization
    progress_bar.progress(100)
    time.sleep(0.5)  # Small delay for visual feedback

# Clear the initialization messages
st.empty()

# Check if models loaded successfully
if not all([TOKENIZER, MODEL_1, MODEL_2, MODEL_3]):
    st.error("Failed to initialize one or more AI detection models. Please try refreshing the page.")
    st.stop()

# Input area
input_text = st.text_area(
    label="Enter text to analyze:",
    placeholder="Type or paste your content here for AI detection analysis...",
    height=200,
    key="text_input"
)

# Analyze button and output
analyze_button = st.button("Analyze Text", key="analyze_button")
result_placeholder = st.empty()

if analyze_button:
    if input_text and input_text.strip():
        with st.spinner('Analyzing text...'):
            classification_result = classify_text(
                input_text,
                TOKENIZER,
                MODEL_1,
                MODEL_2,
                MODEL_3,
                DEVICE,
                LABEL_MAPPING,
                HUMAN_LABEL_INDEX
            )

        # Display result
        if classification_result is None:
            result_placeholder.warning("Please enter some text to analyze.")
        elif classification_result.get("error"):
            error_message = classification_result.get("message", "An unknown error occurred during analysis.")
            result_placeholder.error(f"Analysis Error: {error_message}")
        elif classification_result["is_human"]:
            prob = classification_result['probability']
            result_html = (
                f"<div class='result-box'>"
                f"<b>The text is</b> <span class='highlight-human'><b>{prob:.2f}%</b> likely <b>Human written</b>.</span>"
                f"</div>"
            )
            result_placeholder.markdown(result_html, unsafe_allow_html=True)
        else:  # AI generated
            prob = classification_result['probability']
            model_name = classification_result['model']
            result_html = (
                f"<div class='result-box'>"
                f"<b>The text is</b> <span class='highlight-ai'><b>{prob:.2f}%</b> likely <b>AI generated</b>.</span><br><br>"
                f"<b>Most Likely AI Model: {model_name}</b>"
                f"</div>"
            )
            result_placeholder.markdown(result_html, unsafe_allow_html=True)
    else:
        result_placeholder.warning("Please enter some text to analyze.")

# Footer
st.markdown("<div class='footer'>Developed by Eeman Majumder</div>", unsafe_allow_html=True)