Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,151 +1,68 @@
|
|
| 1 |
-
import gradio as gr
|
| 2 |
import os
|
| 3 |
-
|
| 4 |
-
|
| 5 |
import numpy as np
|
| 6 |
-
import
|
| 7 |
-
import
|
| 8 |
|
| 9 |
-
#
|
| 10 |
-
|
| 11 |
-
files = {
|
| 12 |
-
"vampires": "vampires.txt"
|
| 13 |
-
}
|
| 14 |
-
|
| 15 |
-
knowledge_base = {}
|
| 16 |
-
for category, filename in files.items():
|
| 17 |
-
try:
|
| 18 |
-
with open(filename, 'r', encoding='utf-8') as file:
|
| 19 |
-
content = file.read()
|
| 20 |
-
# Разбиваем на осмысленные блоки (абзацы)
|
| 21 |
-
paragraphs = [p.strip() for p in content.split('\n\n') if p.strip()]
|
| 22 |
-
knowledge_base[category] = paragraphs
|
| 23 |
-
except FileNotFoundError:
|
| 24 |
-
print(f"Файл {filename} не найден")
|
| 25 |
-
knowledge_base[category] = []
|
| 26 |
-
|
| 27 |
-
return knowledge_base
|
| 28 |
|
| 29 |
-
#
|
| 30 |
-
|
| 31 |
-
return SentenceTransformer('sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2')
|
| 32 |
|
| 33 |
-
#
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
for category, paragraphs in knowledge_base.items():
|
| 37 |
-
for para in paragraphs:
|
| 38 |
-
all_fragments.append((para, category))
|
| 39 |
-
|
| 40 |
-
if not all_fragments:
|
| 41 |
-
return []
|
| 42 |
-
|
| 43 |
-
texts = [f[0] for f in all_fragments]
|
| 44 |
-
embeddings = model.encode(texts)
|
| 45 |
-
question_embedding = model.encode([question])
|
| 46 |
-
|
| 47 |
-
similarities = np.dot(embeddings, question_embedding.T).flatten()
|
| 48 |
-
top_indices = similarities.argsort()[-top_k:][::-1]
|
| 49 |
-
|
| 50 |
-
return [all_fragments[i] for i in top_indices]
|
| 51 |
|
| 52 |
-
#
|
| 53 |
-
def
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
endings = [
|
| 84 |
-
"Надеюсь, эта информация была полезной!",
|
| 85 |
-
"Если хотите узнать больше деталей, уточните вопрос.",
|
| 86 |
-
"Могу уточнить какие-то моменты, если нужно.",
|
| 87 |
-
"Это основные сведения, которые у меня есть."
|
| 88 |
-
]
|
| 89 |
-
|
| 90 |
-
response += random.choice(endings)
|
| 91 |
-
|
| 92 |
return response
|
| 93 |
|
| 94 |
-
#
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
if detect(question) != 'ru':
|
| 98 |
-
return "Пожалуйста, задавайте вопросы на русском языке.", history
|
| 99 |
-
except:
|
| 100 |
-
pass
|
| 101 |
-
|
| 102 |
-
if not hasattr(process_question, 'knowledge_base'):
|
| 103 |
-
process_question.knowledge_base = load_and_preprocess_files()
|
| 104 |
-
|
| 105 |
-
if not hasattr(process_question, 'search_model'):
|
| 106 |
-
process_question.search_model = initialize_search_model()
|
| 107 |
-
|
| 108 |
-
relevant_info = find_relevant_info(question, process_question.knowledge_base, process_question.search_model)
|
| 109 |
-
answer = generate_natural_response(question, relevant_info)
|
| 110 |
-
history.append((question, answer))
|
| 111 |
-
return "", history
|
| 112 |
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
gr.Markdown("""<h1 style='text-align: center'>🧛♂️ Мир сверхъестественного 🐺</h1>""")
|
| 116 |
-
gr.Markdown("""<div style='text-align: center'>Задавайте вопросы о вампирах, оборотнях и людях на русском языке</div>""")
|
| 117 |
-
|
| 118 |
-
# Сначала определяем элементы ввода
|
| 119 |
-
msg = gr.Textbox(
|
| 120 |
-
label="Ваш вопрос",
|
| 121 |
-
placeholder="Введите вопрос и нажмите Enter...",
|
| 122 |
-
container=False
|
| 123 |
-
)
|
| 124 |
-
|
| 125 |
-
# Затем определяем примеры, которые используют msg
|
| 126 |
-
examples = gr.Examples(
|
| 127 |
examples=[
|
| 128 |
-
"Какие
|
| 129 |
-
"
|
| 130 |
-
"
|
| 131 |
-
"Расскажи подробнее о вампирах"
|
| 132 |
],
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
)
|
| 136 |
-
|
| 137 |
-
# Затем определяем чат
|
| 138 |
-
chatbot = gr.Chatbot(
|
| 139 |
-
label="Диалог",
|
| 140 |
-
height=500
|
| 141 |
)
|
| 142 |
-
|
| 143 |
-
with gr.Row():
|
| 144 |
-
submit = gr.Button("Отправить", variant="primary")
|
| 145 |
-
clear = gr.Button("Очистить историю")
|
| 146 |
-
|
| 147 |
-
submit.click(process_question, [msg, chatbot], [msg, chatbot])
|
| 148 |
-
msg.submit(process_question, [msg, chatbot], [msg, chatbot])
|
| 149 |
-
clear.click(lambda: None, None, chatbot, queue=False)
|
| 150 |
|
| 151 |
-
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
+
import glob
|
| 3 |
+
import gradio as gr
|
| 4 |
import numpy as np
|
| 5 |
+
from sentence_transformers import SentenceTransformer
|
| 6 |
+
from sklearn.metrics.pairwise import cosine_similarity
|
| 7 |
|
| 8 |
+
# Загружаем модель эмбеддингов
|
| 9 |
+
model = SentenceTransformer('sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
|
| 11 |
+
# Путь к папке с лором
|
| 12 |
+
LORE_DIR = "./lore"
|
|
|
|
| 13 |
|
| 14 |
+
# Параметры нарезки текста
|
| 15 |
+
CHUNK_SIZE = 500 # длина куска в символах
|
| 16 |
+
CHUNK_OVERLAP = 100 # перекрытие для плавности
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
+
# Загружаем и разбиваем текст на кусочки
|
| 19 |
+
def load_lore_chunks():
|
| 20 |
+
chunks = []
|
| 21 |
+
files = glob.glob(os.path.join(LORE_DIR, "*.txt"))
|
| 22 |
+
if not files:
|
| 23 |
+
print(f"В папке {LORE_DIR} нет файлов.")
|
| 24 |
+
for file_path in files:
|
| 25 |
+
with open(file_path, "r", encoding="utf-8") as file:
|
| 26 |
+
text = file.read()
|
| 27 |
+
text = ''.join(c if 0x20 <= ord(c) <= 0xFFFF else ' ' for c in text) # чистим мусор
|
| 28 |
+
for i in range(0, len(text), CHUNK_SIZE - CHUNK_OVERLAP):
|
| 29 |
+
chunk = text[i:i+CHUNK_SIZE].strip()
|
| 30 |
+
if chunk:
|
| 31 |
+
chunks.append(chunk)
|
| 32 |
+
return chunks
|
| 33 |
+
|
| 34 |
+
# Загружаем все куски и эмбеддим
|
| 35 |
+
print("🚀 Загружаем лор...")
|
| 36 |
+
lore_chunks = load_lore_chunks()
|
| 37 |
+
if not lore_chunks:
|
| 38 |
+
print("⚠️ Внимание: нет данных для поиска.")
|
| 39 |
+
lore_embeddings = model.encode(lore_chunks)
|
| 40 |
+
print(f"✅ Загружено {len(lore_chunks)} кусков текста.")
|
| 41 |
+
|
| 42 |
+
# Поиск лучшего ответа
|
| 43 |
+
def find_best_answer(question):
|
| 44 |
+
question_embedding = model.encode([question])[0]
|
| 45 |
+
similarities = cosine_similarity([question_embedding], lore_embeddings)[0]
|
| 46 |
+
best_indices = similarities.argsort()[-3:][::-1] # Топ-3 самых похожих
|
| 47 |
+
best_chunks = [lore_chunks[idx] for idx in best_indices]
|
| 48 |
+
response = "\n\n".join(best_chunks)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
return response
|
| 50 |
|
| 51 |
+
# Gradio интерфейс
|
| 52 |
+
with gr.Blocks() as demo:
|
| 53 |
+
gr.Markdown("## 🧛♂️ ЛОР-БОТ: задавай вопросы о мире!")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
|
| 55 |
+
chat = gr.ChatInterface(
|
| 56 |
+
fn=lambda message, history: (find_best_answer(message), history),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
examples=[
|
| 58 |
+
["Какие кланы есть у вампиров?"],
|
| 59 |
+
["Чем оборотни отличаются от ликантропов?"],
|
| 60 |
+
["Где находится замок теней?"]
|
|
|
|
| 61 |
],
|
| 62 |
+
title="Лор-бот",
|
| 63 |
+
theme="soft"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 64 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
|
| 66 |
+
# Для запуска локально:
|
| 67 |
+
if __name__ == "__main__":
|
| 68 |
+
demo.launch()
|