Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,15 +1,18 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
from transformers import pipeline, AutoTokenizer
|
| 3 |
-
import threading
|
|
|
|
| 4 |
from langdetect import detect
|
| 5 |
|
|
|
|
|
|
|
| 6 |
models = {}
|
| 7 |
models_ready = False
|
|
|
|
| 8 |
|
| 9 |
def load_model_safe(model_name, task="summarization"):
|
| 10 |
try:
|
| 11 |
print(f"🔄 Загружается модель: {model_name}")
|
| 12 |
-
# Принудительно используем legacy=False только если доступен fast tokenizer
|
| 13 |
tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False, legacy=True)
|
| 14 |
model = pipeline(task, model=model_name, tokenizer=tokenizer)
|
| 15 |
return model
|
|
@@ -18,16 +21,16 @@ def load_model_safe(model_name, task="summarization"):
|
|
| 18 |
return None
|
| 19 |
|
| 20 |
def preload_models():
|
| 21 |
-
global models, models_ready
|
| 22 |
try:
|
| 23 |
-
print("🚀 Инициализация моделей...")
|
| 24 |
models["en"] = load_model_safe("facebook/bart-large-cnn")
|
| 25 |
models["ru"] = load_model_safe("IlyaGusev/mbart_ru_sum_gazeta")
|
| 26 |
models["multi"] = load_model_safe("csebuetnlp/mT5_multilingual_XLSum")
|
| 27 |
models_ready = True
|
| 28 |
-
|
|
|
|
| 29 |
except Exception as e:
|
| 30 |
-
|
| 31 |
|
| 32 |
threading.Thread(target=preload_models, daemon=True).start()
|
| 33 |
|
|
@@ -35,7 +38,7 @@ def summarize_text(text):
|
|
| 35 |
if not text.strip():
|
| 36 |
return "⚠️ Пожалуйста, введите текст."
|
| 37 |
if not models_ready:
|
| 38 |
-
return
|
| 39 |
try:
|
| 40 |
lang = detect(text)
|
| 41 |
if lang == "ru":
|
|
@@ -49,11 +52,12 @@ def summarize_text(text):
|
|
| 49 |
except Exception as e:
|
| 50 |
return f"⚠️ Ошибка обработки: {e}"
|
| 51 |
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
)
|
|
|
|
| 59 |
iface.launch(server_name="0.0.0.0", server_port=7860)
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
from transformers import pipeline, AutoTokenizer
|
| 3 |
+
import threading
|
| 4 |
+
import warnings
|
| 5 |
from langdetect import detect
|
| 6 |
|
| 7 |
+
warnings.filterwarnings("ignore")
|
| 8 |
+
|
| 9 |
models = {}
|
| 10 |
models_ready = False
|
| 11 |
+
init_message = "⏳ Инициализация моделей (1–2 мин)..."
|
| 12 |
|
| 13 |
def load_model_safe(model_name, task="summarization"):
|
| 14 |
try:
|
| 15 |
print(f"🔄 Загружается модель: {model_name}")
|
|
|
|
| 16 |
tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False, legacy=True)
|
| 17 |
model = pipeline(task, model=model_name, tokenizer=tokenizer)
|
| 18 |
return model
|
|
|
|
| 21 |
return None
|
| 22 |
|
| 23 |
def preload_models():
|
| 24 |
+
global models, models_ready, init_message
|
| 25 |
try:
|
|
|
|
| 26 |
models["en"] = load_model_safe("facebook/bart-large-cnn")
|
| 27 |
models["ru"] = load_model_safe("IlyaGusev/mbart_ru_sum_gazeta")
|
| 28 |
models["multi"] = load_model_safe("csebuetnlp/mT5_multilingual_XLSum")
|
| 29 |
models_ready = True
|
| 30 |
+
init_message = "✅ Все модели успешно загружены!"
|
| 31 |
+
print(init_message)
|
| 32 |
except Exception as e:
|
| 33 |
+
init_message = f"❌ Ошибка инициализации моделей: {e}"
|
| 34 |
|
| 35 |
threading.Thread(target=preload_models, daemon=True).start()
|
| 36 |
|
|
|
|
| 38 |
if not text.strip():
|
| 39 |
return "⚠️ Пожалуйста, введите текст."
|
| 40 |
if not models_ready:
|
| 41 |
+
return init_message
|
| 42 |
try:
|
| 43 |
lang = detect(text)
|
| 44 |
if lang == "ru":
|
|
|
|
| 52 |
except Exception as e:
|
| 53 |
return f"⚠️ Ошибка обработки: {e}"
|
| 54 |
|
| 55 |
+
with gr.Blocks(title="Eroha Summarizer PRO") as iface:
|
| 56 |
+
gr.Markdown("## 🧠 **Eroha Summarizer PRO**\nМногоязычный суммаризатор на BART / MBART / mT5\n")
|
| 57 |
+
gr.Markdown(init_message)
|
| 58 |
+
inp = gr.Textbox(label="Введите текст для суммаризации", lines=12)
|
| 59 |
+
out = gr.Textbox(label="Результат суммаризации")
|
| 60 |
+
btn = gr.Button("🚀 Суммаризировать")
|
| 61 |
+
btn.click(summarize_text, inputs=inp, outputs=out)
|
| 62 |
+
|
| 63 |
iface.launch(server_name="0.0.0.0", server_port=7860)
|