Upload 2 files
Browse files- app.py +77 -0
- requirements.txt +5 -0
app.py
ADDED
|
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 3 |
+
import torch
|
| 4 |
+
|
| 5 |
+
# Загрузка модели SmolLM3-3B
|
| 6 |
+
print("🚀 Загрузка SmolLM3-3B...")
|
| 7 |
+
model_name = "HuggingFaceTB/SmolLM3-3B"
|
| 8 |
+
|
| 9 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 10 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 11 |
+
model_name,
|
| 12 |
+
torch_dtype=torch.float32,
|
| 13 |
+
low_cpu_mem_usage=True
|
| 14 |
+
)
|
| 15 |
+
|
| 16 |
+
print("✅ Модель загружена!")
|
| 17 |
+
|
| 18 |
+
def chat(message, history):
|
| 19 |
+
"""Обработка сообщений в чате"""
|
| 20 |
+
# Собираем историю разговора
|
| 21 |
+
conversation = ""
|
| 22 |
+
for user_msg, bot_msg in history:
|
| 23 |
+
conversation += f"User: {user_msg}\nAssistant: {bot_msg}\n"
|
| 24 |
+
conversation += f"User: {message}\nAssistant:"
|
| 25 |
+
|
| 26 |
+
# Генерируем ответ
|
| 27 |
+
inputs = tokenizer(conversation, return_tensors="pt", truncation=True, max_length=2048)
|
| 28 |
+
|
| 29 |
+
with torch.no_grad():
|
| 30 |
+
outputs = model.generate(
|
| 31 |
+
**inputs,
|
| 32 |
+
max_new_tokens=300,
|
| 33 |
+
temperature=0.7,
|
| 34 |
+
do_sample=True,
|
| 35 |
+
top_p=0.9,
|
| 36 |
+
repetition_penalty=1.1
|
| 37 |
+
)
|
| 38 |
+
|
| 39 |
+
# Декодируем ответ
|
| 40 |
+
full_response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 41 |
+
|
| 42 |
+
# Извлекаем только новый ответ ассистента
|
| 43 |
+
if "Assistant:" in full_response:
|
| 44 |
+
response = full_response.split("Assistant:")[-1].strip()
|
| 45 |
+
else:
|
| 46 |
+
response = full_response.strip()
|
| 47 |
+
|
| 48 |
+
return response
|
| 49 |
+
|
| 50 |
+
# Создаём интерфейс
|
| 51 |
+
demo = gr.ChatInterface(
|
| 52 |
+
fn=chat,
|
| 53 |
+
title="🌍 SmolLM3-3B Multilingual Chat",
|
| 54 |
+
description="""
|
| 55 |
+
**Мультиязычный AI чат-бот на базе SmolLM3-3B (3 миллиарда параметров)**
|
| 56 |
+
|
| 57 |
+
Поддерживаемые языки:
|
| 58 |
+
🇬🇧 English | 🇷🇺 Русский | 🇪🇸 Español | 🇫🇷 Français | 🇩🇪 Deutsch | 🇮🇹 Italiano | 🇵🇹 Português
|
| 59 |
+
|
| 60 |
+
Полностью бесплатный и без ограничений! 🚀
|
| 61 |
+
""",
|
| 62 |
+
examples=[
|
| 63 |
+
"Привет! Расскажи о себе",
|
| 64 |
+
"Hello! Tell me about yourself",
|
| 65 |
+
"¿Hola! ¿Qué puedes hacer?",
|
| 66 |
+
"Bonjour! Que peux-tu faire?",
|
| 67 |
+
"Hallo! Was kannst du tun?",
|
| 68 |
+
],
|
| 69 |
+
cache_examples=False,
|
| 70 |
+
retry_btn="🔄 Повторить",
|
| 71 |
+
undo_btn="↩️ Отменить",
|
| 72 |
+
clear_btn="🗑️ Очистить",
|
| 73 |
+
)
|
| 74 |
+
|
| 75 |
+
# Запуск
|
| 76 |
+
if __name__ == "__main__":
|
| 77 |
+
demo.launch()
|
requirements.txt
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
transformers>=4.35.0
|
| 2 |
+
torch>=2.0.0
|
| 3 |
+
gradio>=4.0.0
|
| 4 |
+
accelerate>=0.24.0
|
| 5 |
+
sentencepiece>=0.1.99
|