Update app.py

app.py

@@ -1,87 +1,45 @@
+from fastapi import FastAPI, UploadFile, File
+from typing import List
 import gradio as gr
-from transformers import pipeline, AutoTokenizer, logging
-from langdetect import detect
-import threading, warnings, os, torch
-
-# --- Silent mode ---
-os.environ["TOKENIZERS_PARALLELISM"] = "false"
-os.environ["TRANSFORMERS_NO_ADVISORY_WARNINGS"] = "1"
-warnings.filterwarnings("ignore")
-logging.set_verbosity_error()
-
-# --- Global vars ---
-models = {}
-models_ready = False
-status_message = "⏳ Инициализация моделей (1–2 мин)..."
-
-# --- Model loader ---
-def load_model(model_name, task="summarization"):
-    try:
-        print(f"🔄 Загружается модель: {model_name}")
-        tokenizer = AutoTokenizer.from_pretrained(
-            model_name,
-            use_fast=False,
-            legacy=True,
-            trust_remote_code=True
-        )
-        model_pipe = pipeline(task, model=model_name, tokenizer=tokenizer, device=-1)
-        # Torch optimizations
-        if hasattr(model_pipe.model, "half"):
-            model_pipe.model = model_pipe.model.half()
-        if hasattr(torch, "compile"):
-            try:
-                model_pipe.model = torch.compile(model_pipe.model, mode="reduce-overhead")
-            except Exception:
-                pass
-        return model_pipe
-    except Exception as e:
-        print(f"⚠️ Ошибка при загрузке {model_name}: {e}")
-        return None
-
-# --- Preload all models asynchronously ---
-def preload_models():
-    global models, models_ready, status_message
-    try:
-        models["en"] = load_model("facebook/bart-large-cnn")
-        models["ru"] = load_model("IlyaGusev/mbart_ru_sum_gazeta")
-        models["multi"] = load_model("csebuetnlp/mT5_multilingual_XLSum")
-        models_ready = True
-        status_message = "✅ Все модели успешно загружены и оптимизированы!"
-        print(status_message)
-    except Exception as e:
-        status_message = f"❌ Ошибка инициализации моделей: {e}"
-        print(status_message)
-
-threading.Thread(target=preload_models, daemon=True).start()
-
-# --- Summarization logic ---
-def summarize_text(text):
-    if not text.strip():
-        return "⚠️ Введите текст для суммаризации."
-    if not models_ready:
-        return status_message
-    try:
-        lang = detect(text)
-        model = (
-            models["ru"] if lang == "ru" else
-            models["en"] if lang in ("en", "fr", "de", "es") else
-            models["multi"]
-        )
-        summary = model(text, max_length=250, min_length=40, do_sample=False)
-        return summary[0]["summary_text"]
-    except Exception as e:
-        return f"⚠️ Ошибка суммаризации: {e}"
-
-# --- UI ---
-with gr.Blocks(title="Eroha Summarizer PRO — Multilingual (BART / MBART / mT5)") as app:
-    gr.Markdown("""
-    ## 🧠 **Eroha Summarizer PRO**
-    <div style='color:gray'>Многоязычный суммаризатор (EN/RU/Multilingual)</div>
-    """)
-    status = gr.Markdown(status_message)
-    inp = gr.Textbox(label="Введите текст", lines=12, placeholder="Вставьте текст сюда...")
-    out = gr.Textbox(label="Результат суммаризации")
-    btn = gr.Button("🚀 Суммаризировать")
-    btn.click(fn=summarize_text, inputs=inp, outputs=out)
-
-app.launch(server_name="0.0.0.0", server_port=7860)
+from transformers import pipeline
+
+app = FastAPI(title="Eroha Summarizer PRO", version="3.6")
+
+# Подгружаем модель (ленивая загрузка)
+summarizer = None
+
+@app.get("/")
+def home():
+    return {"message": "✅ Eroha Summarizer PRO работает!"}
+
+@app.post("/summarize")
+async def summarize(files: List[UploadFile] = File(...)):
+    global summarizer
+    if summarizer is None:
+        summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
+    
+    texts = [await f.read() for f in files]
+    full_text = "\n".join([t.decode("utf-8") for t in texts])
+    summary = summarizer(full_text, max_length=180, min_length=50, do_sample=False)
+    return {"summary": summary[0]["summary_text"]}
+
+# ============ Gradio UI ============
+def gradio_summary(text):
+    global summarizer
+    if summarizer is None:
+        summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
+    result = summarizer(text, max_length=180, min_length=50, do_sample=False)
+    return result[0]["summary_text"]
+
+iface = gr.Interface(
+    fn=gradio_summary,
+    inputs=gr.Textbox(lines=10, label="Введите текст для суммаризации"),
+    outputs="text",
+    title="Eroha Summarizer PRO",
+    description="AI-система для анализа и суммаризации текстов."
+)
+
+@app.get("/gradio")
+def launch_gradio():
+    iface.launch(share=False, server_name="0.0.0.0", server_port=7860)
+    return {"message": "Gradio interface запущен!"}