app.py

import os
import gradio as gr
from transformers import pipeline
import pdfplumber
import docx2txt
from docx import Document
from fpdf import FPDF
from langdetect import detect
import urllib.request


# === 🗂️ Инициализация ===
os.makedirs("/app/models", exist_ok=True)
FONT_PATH = "DejaVuSans.ttf"

if not os.path.exists(FONT_PATH):
    print("⬇️ Загружаю шрифт DejaVuSans.ttf ...")
    urllib.request.urlretrieve(
        "https://github.com/dejavu-fonts/dejavu-fonts/raw/master/ttf/DejaVuSans.ttf",
        FONT_PATH
    )


# === ⚙️ Модели ===
def load_model(task, model_name):
    print(f"🔹 Загружается модель: {model_name}")
    return pipeline(task, model=model_name, cache_dir="/app/models")

summarizers = {
    "en": load_model("summarization", "facebook/bart-large-cnn"),
    "ru": load_model("summarization", "IlyaGusev/mbart_ru_sum_gazeta"),
    "kz": load_model("summarization", "csebuetnlp/mT5_multilingual_XLSum")
}


# === 📄 Чтение файлов ===
def read_file(file):
    if not file:
        return ""
    filename = file.name.lower()
    text = ""
    try:
        if filename.endswith(".pdf"):
            with pdfplumber.open(file.name) as pdf:
                for page in pdf.pages:
                    if page.extract_text():
                        text += page.extract_text() + "\n"
        elif filename.endswith(".docx"):
            text = docx2txt.process(file.name)
        else:
            text = file.read().decode("utf-8", errors="ignore")
    except Exception as e:
        return f"⚠️ Ошибка при чтении файла: {e}"
    return text.strip()


# === 🌐 Определение языка ===
def detect_language(text):
    try:
        lang = detect(text)
        if lang.startswith("ru"):
            return "ru"
        elif lang.startswith("kk") or any(x in text for x in "әіңғүұқөһ"):
            return "kz"
        else:
            return "en"
    except:
        return "en"


# === 🧠 Суммаризация ===
def summarize_text(text):
    if not text or len(text) < 50:
        return "⚠️ Недостаточно текста для анализа.", "❌", "❌", 0, 0, "❌"

    lang = detect_language(text)
    model = summarizers.get(lang, summarizers["en"])
    flags = {"ru": "🇷🇺 Русский", "kz": "🇰🇿 Қазақ тілі", "en": "🇬🇧 English"}
    lang_label = flags.get(lang, "🌍 Unknown")
    model_label = model.model.name_or_path if hasattr(model.model, "name_or_path") else "Custom"

    chunk_size = 2500
    overlap = 200
    summaries = []

    for i in range(0, len(text), chunk_size - overlap):
        chunk = text[i:i + chunk_size]
        try:
            result = model(chunk, max_length=180, min_length=40, do_sample=False)
            summaries.append(result[0]['summary_text'])
        except Exception as e:
            summaries.append(f"[Ошибка в части {len(summaries)+1}: {e}]")

    summary = "\n\n".join(summaries).strip()
    src_len = len(text)
    sum_len = len(summary)
    compression = round(100 * (1 - sum_len / src_len), 1) if src_len > 0 else 0

    return summary, lang_label, model_label, src_len, sum_len, f"{compression}%"


# === 💾 Сохранение ===
def save_summary_as_txt(summary_text):
    path = "summary.txt"
    with open(path, "w", encoding="utf-8") as f:
        f.write(summary_text)
    return path

def save_summary_as_docx(summary_text):
    path = "summary.docx"
    doc = Document()
    doc.add_heading("Резюме документа", level=1)
    doc.add_paragraph(summary_text)
    doc.save(path)
    return path

def save_summary_as_pdf(summary_text):
    path = "summary.pdf"
    pdf = FPDF()
    pdf.add_page()
    pdf.add_font('DejaVu', '', FONT_PATH, uni=True)
    pdf.set_font('DejaVu', '', 12)
    pdf.multi_cell(0, 10, summary_text)
    pdf.output(path)
    return path


# === 🚀 Основная функция ===
def summarize_file(file):
    text = read_file(file)
    if text.startswith("⚠️"):
        return text, "❌", "❌", 0, 0, "❌", None, None, None

    with gr.Progress(track_tqdm=True) as progress:
        progress(0, desc="🧠 Анализ текста...")
        summary, lang_label, model_label, src_len, sum_len, compression = summarize_text(text)
        progress(1, desc="✅ Готово!")

    txt_path = save_summary_as_txt(summary)
    docx_path = save_summary_as_docx(summary)
    pdf_path = save_summary_as_pdf(summary)

    return summary, lang_label, model_label, src_len, sum_len, compression, txt_path, docx_path, pdf_path


# === 🧩 Интерфейс (универсальный для любых версий Gradio 4.4+) ===
with gr.Blocks() as demo:
    gr.Markdown(
        "<h2 style='text-align:center'>🧠 Eroha Summarizer PRO</h2>"
        "<p style='text-align:center'>Определяет язык (🇷🇺 / 🇰🇿 / 🇬🇧), создаёт краткое резюме и сохраняет в TXT, DOCX, PDF</p>"
    )

    file_input = gr.File(label="📂 Загрузите документ (.pdf, .docx, .txt)")
    run_btn = gr.Button("🔍 Сгенерировать резюме", variant="primary")

    summary_output = gr.Textbox(label="🧾 Краткое резюме", lines=10)
    lang_output = gr.Textbox(label="🌍 Определённый язык")
    model_output = gr.Textbox(label="🧠 Используемая модель")
    src_len = gr.Number(label="📄 Длина исходного текста")
    sum_len = gr.Number(label="📝 Длина резюме")
    compression = gr.Textbox(label="📉 Степень сокращения")
    txt_file = gr.File(label="📄 TXT файл")
    docx_file = gr.File(label="📘 DOCX файл")
    pdf_file = gr.File(label="📕 PDF файл")

    run_btn.click(
        summarize_file,
        inputs=[file_input],
        outputs=[
            summary_output, lang_output, model_output,
            src_len, sum_len, compression,
            txt_file, docx_file, pdf_file
        ]
    )

demo.launch(server_name="0.0.0.0", server_port=7860)