Spaces:
Sleeping
Sleeping
Upload 2 files
Browse files- app.py +428 -0
- requirements.txt +0 -0
app.py
ADDED
|
@@ -0,0 +1,428 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from typing import List, Dict, Tuple
|
| 2 |
+
import numpy as np
|
| 3 |
+
from sentence_transformers import SentenceTransformer, CrossEncoder
|
| 4 |
+
from rank_bm25 import BM25Okapi
|
| 5 |
+
from groq import Groq
|
| 6 |
+
import gradio as gr
|
| 7 |
+
from dataclasses import dataclass
|
| 8 |
+
import re
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
@dataclass
|
| 12 |
+
class Chunk:
|
| 13 |
+
id: int
|
| 14 |
+
text: str
|
| 15 |
+
position: int
|
| 16 |
+
metadata: Dict = None
|
| 17 |
+
|
| 18 |
+
def __post_init__(self):
|
| 19 |
+
if self.metadata is None:
|
| 20 |
+
self.metadata = {}
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
class DocumentChunker:
|
| 24 |
+
def __init__(self, chunk_size: int = 500, overlap: int = 100):
|
| 25 |
+
self.chunk_size = chunk_size
|
| 26 |
+
self.overlap = overlap
|
| 27 |
+
|
| 28 |
+
def chunk_text(self, text: str) -> List[Chunk]:
|
| 29 |
+
# Розбиття на речення
|
| 30 |
+
sentences = re.split(r'[.!?]+', text)
|
| 31 |
+
sentences = [s.strip() for s in sentences if s.strip()]
|
| 32 |
+
|
| 33 |
+
chunks = []
|
| 34 |
+
current_chunk = ""
|
| 35 |
+
chunk_id = 0
|
| 36 |
+
|
| 37 |
+
for sentence in sentences:
|
| 38 |
+
if len(current_chunk) + len(sentence) > self.chunk_size and current_chunk:
|
| 39 |
+
chunks.append(Chunk(
|
| 40 |
+
id=chunk_id,
|
| 41 |
+
text=current_chunk.strip(),
|
| 42 |
+
position=chunk_id,
|
| 43 |
+
metadata={'sentence_count': len(current_chunk.split('.'))}
|
| 44 |
+
))
|
| 45 |
+
|
| 46 |
+
# Створення overlap
|
| 47 |
+
words = current_chunk.split()
|
| 48 |
+
overlap_words = words[-int(self.overlap / 5):] if len(words) > int(self.overlap / 5) else words
|
| 49 |
+
current_chunk = ' '.join(overlap_words) + ' ' + sentence
|
| 50 |
+
chunk_id += 1
|
| 51 |
+
else:
|
| 52 |
+
current_chunk += ' ' + sentence
|
| 53 |
+
|
| 54 |
+
# Додавання останнього чанка
|
| 55 |
+
if current_chunk.strip():
|
| 56 |
+
chunks.append(Chunk(
|
| 57 |
+
id=chunk_id,
|
| 58 |
+
text=current_chunk.strip(),
|
| 59 |
+
position=chunk_id,
|
| 60 |
+
metadata={'sentence_count': len(current_chunk.split('.'))}
|
| 61 |
+
))
|
| 62 |
+
|
| 63 |
+
return chunks
|
| 64 |
+
|
| 65 |
+
|
| 66 |
+
class HybridRetriever:
|
| 67 |
+
"""Гібридний retriever з BM25 та semantic search"""
|
| 68 |
+
|
| 69 |
+
def __init__(self, model_name: str = 'sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2'):
|
| 70 |
+
self.embedding_model = SentenceTransformer(model_name)
|
| 71 |
+
self.bm25 = None
|
| 72 |
+
self.chunks = []
|
| 73 |
+
self.embeddings = None
|
| 74 |
+
self.tokenized_corpus = []
|
| 75 |
+
|
| 76 |
+
def index_documents(self, chunks: List[Chunk]):
|
| 77 |
+
self.chunks = chunks
|
| 78 |
+
texts = [chunk.text for chunk in chunks]
|
| 79 |
+
|
| 80 |
+
self.tokenized_corpus = [self._tokenize(text) for text in texts]
|
| 81 |
+
self.bm25 = BM25Okapi(self.tokenized_corpus)
|
| 82 |
+
|
| 83 |
+
print("Створення embeddings...")
|
| 84 |
+
self.embeddings = self.embedding_model.encode(texts, show_progress_bar=True)
|
| 85 |
+
|
| 86 |
+
def _tokenize(self, text: str) -> List[str]:
|
| 87 |
+
text = text.lower()
|
| 88 |
+
text = re.sub(r'[^\wа-яїієґ\s]', ' ', text)
|
| 89 |
+
tokens = text.split()
|
| 90 |
+
return [t for t in tokens if len(t) > 2]
|
| 91 |
+
|
| 92 |
+
def bm25_search(self, query: str, top_k: int = 10) -> List[Tuple[Chunk, float]]:
|
| 93 |
+
if self.bm25 is None:
|
| 94 |
+
return []
|
| 95 |
+
|
| 96 |
+
tokenized_query = self._tokenize(query)
|
| 97 |
+
scores = self.bm25.get_scores(tokenized_query)
|
| 98 |
+
|
| 99 |
+
top_indices = np.argsort(scores)[::-1][:top_k]
|
| 100 |
+
results = [(self.chunks[i], scores[i]) for i in top_indices]
|
| 101 |
+
return results
|
| 102 |
+
|
| 103 |
+
def semantic_search(self, query: str, top_k: int = 10) -> List[Tuple[Chunk, float]]:
|
| 104 |
+
if self.embeddings is None:
|
| 105 |
+
return []
|
| 106 |
+
|
| 107 |
+
query_embedding = self.embedding_model.encode([query])[0]
|
| 108 |
+
|
| 109 |
+
# Косинусна подібність
|
| 110 |
+
similarities = np.dot(self.embeddings, query_embedding) / (
|
| 111 |
+
np.linalg.norm(self.embeddings, axis=1) * np.linalg.norm(query_embedding)
|
| 112 |
+
)
|
| 113 |
+
|
| 114 |
+
top_indices = np.argsort(similarities)[::-1][:top_k]
|
| 115 |
+
results = [(self.chunks[i], similarities[i]) for i in top_indices]
|
| 116 |
+
return results
|
| 117 |
+
|
| 118 |
+
def hybrid_search(self, query: str, top_k: int = 10,
|
| 119 |
+
alpha: float = 0.5) -> List[Tuple[Chunk, float]]:
|
| 120 |
+
bm25_results = self.bm25_search(query, top_k * 2)
|
| 121 |
+
semantic_results = self.semantic_search(query, top_k * 2)
|
| 122 |
+
|
| 123 |
+
bm25_scores = {chunk.id: score for chunk, score in bm25_results}
|
| 124 |
+
semantic_scores = {chunk.id: score for chunk, score in semantic_results}
|
| 125 |
+
|
| 126 |
+
combined_scores = {}
|
| 127 |
+
all_ids = set(bm25_scores.keys()) | set(semantic_scores.keys())
|
| 128 |
+
|
| 129 |
+
for chunk_id in all_ids:
|
| 130 |
+
bm25_score = bm25_scores.get(chunk_id, 0)
|
| 131 |
+
semantic_score = semantic_scores.get(chunk_id, 0)
|
| 132 |
+
|
| 133 |
+
if bm25_results:
|
| 134 |
+
max_bm25 = max(bm25_scores.values())
|
| 135 |
+
bm25_score = bm25_score / max_bm25 if max_bm25 > 0 else 0
|
| 136 |
+
|
| 137 |
+
combined_scores[chunk_id] = alpha * bm25_score + (1 - alpha) * semantic_score
|
| 138 |
+
|
| 139 |
+
sorted_ids = sorted(combined_scores.items(), key=lambda x: x[1], reverse=True)[:top_k]
|
| 140 |
+
results = [(next(c for c in self.chunks if c.id == cid), score)
|
| 141 |
+
for cid, score in sorted_ids]
|
| 142 |
+
return results
|
| 143 |
+
|
| 144 |
+
|
| 145 |
+
class Reranker:
|
| 146 |
+
def __init__(self, model_name: str = 'cross-encoder/ms-marco-MiniLM-L-6-v2'):
|
| 147 |
+
self.model = CrossEncoder(model_name)
|
| 148 |
+
|
| 149 |
+
def rerank(self, query: str, chunks: List[Tuple[Chunk, float]],
|
| 150 |
+
top_k: int = 5) -> List[Tuple[Chunk, float]]:
|
| 151 |
+
if not chunks:
|
| 152 |
+
return []
|
| 153 |
+
|
| 154 |
+
pairs = [[query, chunk.text] for chunk, _ in chunks]
|
| 155 |
+
|
| 156 |
+
scores = self.model.predict(pairs)
|
| 157 |
+
|
| 158 |
+
results = list(zip([chunk for chunk, _ in chunks], scores))
|
| 159 |
+
results.sort(key=lambda x: x[1], reverse=True)
|
| 160 |
+
return results[:top_k]
|
| 161 |
+
|
| 162 |
+
|
| 163 |
+
class RAGSystem:
|
| 164 |
+
def __init__(self, api_key: str = None, model: str = "llama-3.3-70b-versatile"):
|
| 165 |
+
self.chunker = DocumentChunker()
|
| 166 |
+
self.retriever = HybridRetriever()
|
| 167 |
+
self.reranker = Reranker()
|
| 168 |
+
self.client = Groq(api_key=api_key) if api_key else None
|
| 169 |
+
self.model = model
|
| 170 |
+
self.chunks = []
|
| 171 |
+
|
| 172 |
+
def load_document(self, text: str) -> str:
|
| 173 |
+
# Chunking
|
| 174 |
+
self.chunks = self.chunker.chunk_text(text)
|
| 175 |
+
|
| 176 |
+
# Індексація
|
| 177 |
+
self.retriever.index_documents(self.chunks)
|
| 178 |
+
|
| 179 |
+
return f"Документ успішно завантажено. Створено {len(self.chunks)} чанків."
|
| 180 |
+
|
| 181 |
+
def answer_question(self, question: str, retrieval_method: str = "hybrid",
|
| 182 |
+
use_reranker: bool = True, show_citations: bool = True) -> Tuple[str, List[Dict]]:
|
| 183 |
+
"""Відповідь на запитання"""
|
| 184 |
+
if not self.chunks:
|
| 185 |
+
return "Спочатку завантажте документ!", []
|
| 186 |
+
|
| 187 |
+
# Retrieval
|
| 188 |
+
if retrieval_method == "bm25":
|
| 189 |
+
retrieved = self.retriever.bm25_search(question, top_k=10)
|
| 190 |
+
elif retrieval_method == "semantic":
|
| 191 |
+
retrieved = self.retriever.semantic_search(question, top_k=10)
|
| 192 |
+
else: # hybrid
|
| 193 |
+
retrieved = self.retriever.hybrid_search(question, top_k=10)
|
| 194 |
+
|
| 195 |
+
# Reranking
|
| 196 |
+
if use_reranker and retrieved:
|
| 197 |
+
retrieved = self.reranker.rerank(question, retrieved, top_k=5)
|
| 198 |
+
|
| 199 |
+
# Генерація відповіді з Groq
|
| 200 |
+
if self.client is None:
|
| 201 |
+
return "API ключ не налаштовано!", []
|
| 202 |
+
|
| 203 |
+
context = "\n\n".join([
|
| 204 |
+
f"[{i + 1}] {chunk.text}"
|
| 205 |
+
for i, (chunk, _) in enumerate(retrieved)
|
| 206 |
+
])
|
| 207 |
+
|
| 208 |
+
prompt = f"""На основі наведеного контексту дайте відповідь на запитання.
|
| 209 |
+
Обов'язково вказуйте номери джерел у квадратних дужках [1], [2] тощо.
|
| 210 |
+
|
| 211 |
+
Контекст:
|
| 212 |
+
{context}
|
| 213 |
+
|
| 214 |
+
Запитання: {question}
|
| 215 |
+
|
| 216 |
+
Відповідь (з цитуванням):"""
|
| 217 |
+
|
| 218 |
+
try:
|
| 219 |
+
chat_completion = self.client.chat.completions.create(
|
| 220 |
+
messages=[
|
| 221 |
+
{
|
| 222 |
+
"role": "system",
|
| 223 |
+
"content": "Ви - помічник, який відповідає на запитання на основі наданого контексту. Завжди цитуйте джерела у квадратних дужках."
|
| 224 |
+
},
|
| 225 |
+
{
|
| 226 |
+
"role": "user",
|
| 227 |
+
"content": prompt
|
| 228 |
+
}
|
| 229 |
+
],
|
| 230 |
+
model=self.model,
|
| 231 |
+
temperature=0.3,
|
| 232 |
+
max_tokens=2048,
|
| 233 |
+
)
|
| 234 |
+
|
| 235 |
+
answer = chat_completion.choices[0].message.content
|
| 236 |
+
|
| 237 |
+
except Exception as e:
|
| 238 |
+
return f"Помилка при генерації відповіді: {str(e)}", []
|
| 239 |
+
|
| 240 |
+
# Формуємо цитування
|
| 241 |
+
citations = []
|
| 242 |
+
if show_citations:
|
| 243 |
+
citations = [
|
| 244 |
+
{"id": i + 1, "text": chunk.text, "score": float(score)}
|
| 245 |
+
for i, (chunk, score) in enumerate(retrieved)
|
| 246 |
+
]
|
| 247 |
+
|
| 248 |
+
return answer, citations
|
| 249 |
+
|
| 250 |
+
|
| 251 |
+
def create_gradio_interface():
|
| 252 |
+
rag_system = None
|
| 253 |
+
|
| 254 |
+
def load_file(file, api_key, model):
|
| 255 |
+
nonlocal rag_system
|
| 256 |
+
|
| 257 |
+
if not api_key:
|
| 258 |
+
return "Введіть Groq API ключ!", "", ""
|
| 259 |
+
|
| 260 |
+
try:
|
| 261 |
+
filename = file.name.lower()
|
| 262 |
+
|
| 263 |
+
if filename.endswith(".txt"):
|
| 264 |
+
with open(file.name, 'r', encoding='utf-8') as f:
|
| 265 |
+
text = f.read()
|
| 266 |
+
|
| 267 |
+
elif filename.endswith(".pdf"):
|
| 268 |
+
import pdfplumber
|
| 269 |
+
text = ""
|
| 270 |
+
with pdfplumber.open(file.name) as pdf:
|
| 271 |
+
for page in pdf.pages:
|
| 272 |
+
text += page.extract_text() + "\n"
|
| 273 |
+
|
| 274 |
+
else:
|
| 275 |
+
return "Формат файлу не підтримується! Завантажте .txt або .pdf", "", ""
|
| 276 |
+
|
| 277 |
+
rag_system = RAGSystem(api_key=api_key, model=model)
|
| 278 |
+
|
| 279 |
+
status = rag_system.load_document(text)
|
| 280 |
+
return status, "", ""
|
| 281 |
+
|
| 282 |
+
except Exception as e:
|
| 283 |
+
return f"Помилка завантаження файлу: {str(e)}", "", ""
|
| 284 |
+
|
| 285 |
+
def answer(question, retrieval_method, use_reranker, show_citations):
|
| 286 |
+
if rag_system is None:
|
| 287 |
+
return "Спочатку завантажте документ!", ""
|
| 288 |
+
|
| 289 |
+
try:
|
| 290 |
+
answer_text, citations = rag_system.answer_question(
|
| 291 |
+
question,
|
| 292 |
+
retrieval_method.lower().replace(" ", ""),
|
| 293 |
+
use_reranker,
|
| 294 |
+
show_citations
|
| 295 |
+
)
|
| 296 |
+
|
| 297 |
+
citations_text = ""
|
| 298 |
+
if citations and show_citations:
|
| 299 |
+
citations_text = "\n\n📚 Джерела:\n\n"
|
| 300 |
+
for cit in citations:
|
| 301 |
+
citations_text += f"[{cit['id']}] {cit['text'][:200]}...\n"
|
| 302 |
+
citations_text += f"Score: {cit['score']:.3f}\n\n"
|
| 303 |
+
|
| 304 |
+
return answer_text, citations_text
|
| 305 |
+
except Exception as e:
|
| 306 |
+
return f"Помилка: {str(e)}", ""
|
| 307 |
+
|
| 308 |
+
# Створення інтерфейсу
|
| 309 |
+
with gr.Blocks() as demo:
|
| 310 |
+
gr.Markdown("""
|
| 311 |
+
# ⚡ RAG Question Answering System з Groq API
|
| 312 |
+
|
| 313 |
+
Швидка система для відповідей на запитання з використанням RAG підходу та Groq LLMs.
|
| 314 |
+
Завантажте українську книгу та отримайте відповіді на свої запитання!
|
| 315 |
+
""")
|
| 316 |
+
|
| 317 |
+
with gr.Row():
|
| 318 |
+
with gr.Column(scale=2):
|
| 319 |
+
api_key_input = gr.Textbox(
|
| 320 |
+
label="🔑 Groq API Key",
|
| 321 |
+
type="password",
|
| 322 |
+
placeholder="gsk_...",
|
| 323 |
+
info="Отримайте безкоштовний ключ на console.groq.com"
|
| 324 |
+
)
|
| 325 |
+
|
| 326 |
+
model_select = gr.Dropdown(
|
| 327 |
+
label="🤖 Модель Groq",
|
| 328 |
+
choices=[
|
| 329 |
+
"llama-3.3-70b-versatile",
|
| 330 |
+
"llama-3.1-70b-versatile",
|
| 331 |
+
"mixtral-8x7b-32768",
|
| 332 |
+
"gemma2-9b-it"
|
| 333 |
+
],
|
| 334 |
+
value="llama-3.3-70b-versatile",
|
| 335 |
+
info="Llama 3.3 70B рекомендується для кращої якості"
|
| 336 |
+
)
|
| 337 |
+
|
| 338 |
+
file_input = gr.File(
|
| 339 |
+
label="📁 Завантажте книгу (.txt або .pdf)",
|
| 340 |
+
file_types=["text", ".txt", ".pdf"]
|
| 341 |
+
)
|
| 342 |
+
|
| 343 |
+
load_btn = gr.Button("📥 Завантажити документ", variant="primary", size="lg")
|
| 344 |
+
status_output = gr.Textbox(label="Статус", interactive=False)
|
| 345 |
+
|
| 346 |
+
with gr.Column(scale=1):
|
| 347 |
+
gr.Markdown("### ⚙️ Налаштування пошуку")
|
| 348 |
+
|
| 349 |
+
retrieval_method = gr.Radio(
|
| 350 |
+
["BM25", "Semantic", "Hybrid"],
|
| 351 |
+
label="Метод пошуку",
|
| 352 |
+
value="Hybrid",
|
| 353 |
+
info="Hybrid комбінує обидва методи"
|
| 354 |
+
)
|
| 355 |
+
|
| 356 |
+
use_reranker = gr.Checkbox(
|
| 357 |
+
label="Використовувати Reranker",
|
| 358 |
+
value=True,
|
| 359 |
+
info="Покращує точність результатів"
|
| 360 |
+
)
|
| 361 |
+
|
| 362 |
+
show_citations = gr.Checkbox(
|
| 363 |
+
label="Показувати цитування",
|
| 364 |
+
value=True,
|
| 365 |
+
info="Відображає джерела інформації"
|
| 366 |
+
)
|
| 367 |
+
|
| 368 |
+
gr.Markdown("---")
|
| 369 |
+
|
| 370 |
+
question_input = gr.Textbox(
|
| 371 |
+
label="❓ Ваше запитання",
|
| 372 |
+
placeholder="Введіть запитання про книгу...",
|
| 373 |
+
lines=2
|
| 374 |
+
)
|
| 375 |
+
|
| 376 |
+
ask_btn = gr.Button("🔍 Знайти відповідь", variant="primary", size="lg")
|
| 377 |
+
|
| 378 |
+
with gr.Row():
|
| 379 |
+
with gr.Column():
|
| 380 |
+
answer_output = gr.Textbox(
|
| 381 |
+
label="💡 Відповідь",
|
| 382 |
+
lines=10,
|
| 383 |
+
)
|
| 384 |
+
|
| 385 |
+
with gr.Column():
|
| 386 |
+
citations_output = gr.Textbox(
|
| 387 |
+
label="📚 Джерела",
|
| 388 |
+
lines=10,
|
| 389 |
+
)
|
| 390 |
+
|
| 391 |
+
gr.Examples(
|
| 392 |
+
examples=[
|
| 393 |
+
"Про що ця книга?",
|
| 394 |
+
"Хто головний герой?",
|
| 395 |
+
"Що сталося в кінці?",
|
| 396 |
+
"Які основні теми розглядаються?",
|
| 397 |
+
],
|
| 398 |
+
inputs=question_input,
|
| 399 |
+
label="💭 Приклади запитань"
|
| 400 |
+
)
|
| 401 |
+
|
| 402 |
+
load_btn.click(
|
| 403 |
+
load_file,
|
| 404 |
+
inputs=[file_input, api_key_input, model_select],
|
| 405 |
+
outputs=[status_output, answer_output, citations_output]
|
| 406 |
+
)
|
| 407 |
+
|
| 408 |
+
ask_btn.click(
|
| 409 |
+
answer,
|
| 410 |
+
inputs=[question_input, retrieval_method, use_reranker, show_citations],
|
| 411 |
+
outputs=[answer_output, citations_output]
|
| 412 |
+
)
|
| 413 |
+
|
| 414 |
+
question_input.submit(
|
| 415 |
+
answer,
|
| 416 |
+
inputs=[question_input, retrieval_method, use_reranker, show_citations],
|
| 417 |
+
outputs=[answer_output, citations_output]
|
| 418 |
+
)
|
| 419 |
+
|
| 420 |
+
return demo
|
| 421 |
+
|
| 422 |
+
|
| 423 |
+
if __name__ == "__main__":
|
| 424 |
+
print("Запуск RAG системи з Groq API")
|
| 425 |
+
|
| 426 |
+
demo = create_gradio_interface()
|
| 427 |
+
demo.launch(share=True)
|
| 428 |
+
|
requirements.txt
ADDED
|
Binary file (2.93 kB). View file
|
|
|