-
Notifications
You must be signed in to change notification settings - Fork 45
/
Copy pathvector_db.py
408 lines (342 loc) · 14.5 KB
/
vector_db.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
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
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
from __future__ import annotations
# Define the script's usage example
USAGE_EXAMPLE = """
Example usage:
To process input *.txt files at input_path and save the vector db output at output_db:
python create_vector_db.py input_path output_db --chunk_size 100 --chunk_overlap 10
Required arguments:
- input_path: Path to the input dir containing the .txt files
- output_path: Path to the output vector db.
Optional arguments:
- --chunk_size: Size of the chunks (default: None).
- --chunk_overlap: Overlap between chunks (default: None).
"""
import argparse
import logging
import os
import sys
from typing import Any, List, Optional, Set
from langchain.text_splitter import CharacterTextSplitter, RecursiveCharacterTextSplitter
from langchain_community.document_loaders import DirectoryLoader, UnstructuredURLLoader
from langchain_community.vectorstores import FAISS, Chroma, Qdrant
from langchain_milvus import Milvus
vectordb_dir = os.path.dirname(os.path.abspath(__file__))
utils_dir = os.path.abspath(os.path.join(vectordb_dir, '..'))
repo_dir = os.path.abspath(os.path.join(utils_dir, '..'))
sys.path.append(repo_dir)
sys.path.append(utils_dir)
import uuid
from utils.model_wrappers.api_gateway import APIGateway
EMBEDDING_MODEL = 'intfloat/e5-large-v2'
NORMALIZE_EMBEDDINGS = True
VECTORDB_LOG_FILE_NAME = 'vector_db.log'
# Configure the logger
logging.basicConfig(
level=logging.INFO, # Set the logging level (e.g., INFO, DEBUG)
format='%(asctime)s [%(levelname)s] - %(message)s', # Define the log message format
handlers=[
logging.StreamHandler(), # Output logs to the console
logging.FileHandler(VECTORDB_LOG_FILE_NAME),
],
)
# Create a logger object
logger = logging.getLogger(__name__)
class VectorDb:
"""
A class for creating, updating and loading FAISS or Chroma vector databases,
to use them with retrieval augmented generation tasks with langchain
Args:
None
Attributes:
None
Methods:
load_files: Load files from an input directory as langchain documents
get_text_chunks: Get text chunks from a list of documents
get_token_chunks: Get token chunks from a list of documents
create_vector_store: Create a vector store from chunks and an embedding model
load_vdb: load a previous stored vector database
update_vdb: Update an existing vector store with new chunks
create_vdb: Create a vector database from the raw files in a specific input directory
"""
def __init__(self) -> None:
self.collection_id = str(uuid.uuid4())
self.vector_collections: Set[Any] = set()
def load_files(
self,
input_path: str,
recursive: bool = False,
load_txt: bool = True,
load_pdf: bool = False,
urls: Optional[List[Any]] = None,
) -> List[Any]:
"""Load files from input location
Args:
input_path : input location of files
recursive (bool, optional): flag to load files recursively. Defaults to False.
load_txt (bool, optional): flag to load txt files. Defaults to True.
load_pdf (bool, optional): flag to load pdf files. Defaults to False.
urls (list, optional): list of urls to load. Defaults to None.
Returns:
list: list of documents
"""
docs = []
text_loader_kwargs = {'autodetect_encoding': True}
loader: DirectoryLoader | UnstructuredURLLoader
if input_path is not None:
if load_txt:
loader = DirectoryLoader(
input_path, glob='*.txt', recursive=recursive, show_progress=True, loader_kwargs=text_loader_kwargs
)
docs.extend(loader.load())
if load_pdf:
loader = DirectoryLoader(
input_path, glob='*.pdf', recursive=recursive, show_progress=True, loader_kwargs=text_loader_kwargs
)
docs.extend(loader.load())
if urls:
loader = UnstructuredURLLoader(urls=urls)
docs.extend(loader.load())
logger.info(f'Total {len(docs)} files loaded')
return docs
def get_text_chunks(
self, docs: List[Any], chunk_size: int, chunk_overlap: int, meta_data: Optional[List[Any]] = None
) -> List[Any]:
"""Gets text chunks. If metadata is not None, it will create chunks with metadata elements.
Args:
docs (list): list of documents or texts. If no metadata is passed, this parameter is a list of documents.
If metadata is passed, this parameter is a list of texts.
chunk_size (int): chunk size in number of characters
chunk_overlap (int): chunk overlap in number of characters
metadata (list, optional): list of metadata in dictionary format. Defaults to None.
Returns:
list: list of documents
"""
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=chunk_size, chunk_overlap=chunk_overlap, length_function=len
)
if meta_data is None:
logger.info(f'Splitter: splitting documents')
chunks = text_splitter.split_documents(docs)
else:
logger.info(f'Splitter: creating documents with metadata')
chunks = text_splitter.create_documents(docs, meta_data)
logger.info(f'Total {len(chunks)} chunks created')
return chunks
def get_token_chunks(self, docs: List[Any], chunk_size: int, chunk_overlap: int, tokenizer: Any) -> List[Any]:
"""Gets token chunks. If metadata is not None, it will create chunks with metadata elements.
Args:
docs (list): list of documents or texts. If no metadata is passed, this parameter is a list of documents.
If metadata is passed, this parameter is a list of texts.
chunk_size (int): chunk size in number of tokens
chunk_overlap (int): chunk overlap in number of tokens
Returns:
list: list of documents
"""
text_splitter = CharacterTextSplitter.from_huggingface_tokenizer(
tokenizer, chunk_size=chunk_size, chunk_overlap=chunk_overlap
)
logger.info(f'Splitter: splitting documents')
chunks = text_splitter.split_documents(docs)
logger.info(f'Total {len(chunks)} chunks created')
return chunks
def create_vector_store(
self,
chunks: list,
embeddings: Any,
db_type: str,
output_db: Optional[str] = None,
collection_name: Optional[str] = None,
) -> Any:
"""Creates a vector store
Args:
chunks (list): list of chunks
embeddings (HuggingFaceInstructEmbeddings): embedding model
db_type (str): vector db type
output_db (str, optional): output path to save the vector db. Defaults to None.
"""
if collection_name is None:
collection_name = f'collection_{self.collection_id}'
logger.info(f'This is the collection name: {collection_name}')
vector_store: FAISS | Qdrant | Chroma | Milvus
if db_type == 'faiss':
vector_store = FAISS.from_documents(documents=chunks, embedding=embeddings)
if output_db:
vector_store.save_local(output_db)
elif db_type == 'chroma':
if output_db:
vector_store = Chroma()
vector_store.delete_collection()
vector_store = Chroma.from_documents(
documents=chunks, embedding=embeddings, persist_directory=output_db, collection_name=collection_name
)
else:
vector_store = Chroma()
vector_store.delete_collection()
vector_store = Chroma.from_documents(
documents=chunks, embedding=embeddings, collection_name=collection_name
)
self.vector_collections.add(collection_name)
elif db_type == 'qdrant':
if output_db:
vector_store = Qdrant.from_documents(
documents=chunks,
embedding=embeddings,
path=output_db,
collection_name='test_collection',
)
else:
vector_store = Qdrant.from_documents(
documents=chunks,
embedding=embeddings,
collection_name='test_collection',
)
elif db_type == 'milvus':
if output_db:
os.makedirs(output_db, exist_ok=True)
uri = os.path.join(output_db, 'milvus.db')
else:
os.makedirs('./data/tmp', exist_ok=True)
uri = './data/tmp/milvus.db'
vector_store = Milvus.from_documents(
documents=chunks,
embedding=embeddings,
collection_name=collection_name,
connection_args={'uri': uri},
index_params={
'metric_type': 'L2',
'index_type': 'AUTOINDEX',
'params': {},
},
)
logger.info(f'Vector store saved to {output_db}')
return vector_store
def load_vdb(
self,
persist_directory: Optional[str],
embedding_model: Any,
db_type: str = 'chroma',
collection_name: Optional[str] = None,
) -> Any:
vector_store: FAISS | Qdrant | Chroma | Milvus
if db_type == 'faiss':
vector_store = FAISS.load_local(persist_directory, embedding_model, allow_dangerous_deserialization=True) # type: ignore
elif db_type == 'chroma':
if collection_name:
vector_store = Chroma(
persist_directory=persist_directory,
embedding_function=embedding_model,
collection_name=collection_name,
)
else:
vector_store = Chroma(persist_directory=persist_directory, embedding_function=embedding_model)
elif db_type == 'qdrant':
# TODO: Implement Qdrant loading
pass
elif db_type == 'milvus':
if collection_name is None:
collection_name = 'LangChainCollection'
if persist_directory.endswith('.db'):
persist_directory = os.path.dirname(persist_directory)
uri = os.path.join(persist_directory, 'milvus.db')
vector_store = Milvus(
embedding_function=embedding_model,
collection_name=collection_name,
connection_args={
'uri': uri,
},
index_params={
'metric_type': 'L2',
'index_type': 'AUTOINDEX',
'params': {},
},
)
else:
raise ValueError(f'Unsupported database type: {db_type}')
return vector_store
def update_vdb(
self,
chunks: List[Any],
embeddings: Any,
db_type: str,
input_db: Optional[str] = None,
output_db: Optional[str] = None,
) -> Any:
if db_type == 'faiss':
vector_store = FAISS.load_local(input_db, embeddings, allow_dangerous_deserialization=True) # type: ignore
new_vector_store = self.create_vector_store(chunks, embeddings, db_type, None)
vector_store.merge_from(new_vector_store)
if output_db:
vector_store.save_local(output_db)
elif db_type == 'chroma':
# TODO implement update method for chroma
pass
elif db_type == 'qdrant':
# TODO implement update method for qdrant
pass
elif db_type == 'milvus':
# TODO implement update method for milvus
pass
return vector_store
def create_vdb(
self,
input_path: str,
chunk_size: int,
chunk_overlap: int,
db_type: str,
output_db: Optional[str] = None,
recursive: bool = False,
tokenizer: Optional[Any] = None,
load_txt: bool = True,
load_pdf: bool = False,
urls: Optional[List[str]] = None,
embedding_type: str = 'cpu',
batch_size: Optional[int] = None,
bundle: Optional[bool] = None,
select_expert: Optional[str] = None,
) -> Any:
docs = self.load_files(input_path, recursive=recursive, load_txt=load_txt, load_pdf=load_pdf, urls=urls)
if tokenizer is None:
chunks = self.get_text_chunks(docs, chunk_size, chunk_overlap)
else:
chunks = self.get_token_chunks(docs, chunk_size, chunk_overlap, tokenizer)
embeddings = APIGateway.load_embedding_model(
type=embedding_type, batch_size=batch_size, bundle=bundle, select_expert=select_expert
)
vector_store = self.create_vector_store(chunks, embeddings, db_type, output_db)
return vector_store
def dir_path(path: str) -> Any:
if os.path.isdir(path):
return path
else:
raise argparse.ArgumentTypeError(f'readable_dir:{path} is not a valid path')
# Parse the arguments
def parse_arguments() -> Any:
parser = argparse.ArgumentParser(description='Process command line arguments.')
parser.add_argument('-input_path', type=dir_path, help='path to input directory')
parser.add_argument('--chunk_size', type=int, help='chunk size for splitting')
parser.add_argument('--chunk_overlap', type=int, help='chunk overlap for splitting')
parser.add_argument('-output_path', type=dir_path, help='path to input directory')
return parser.parse_args()
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Process data with optional chunking')
# Required arguments
parser.add_argument('--input_path', type=str, help='Path to the input directory')
parser.add_argument('--output_db', type=str, help='Path to the output vectordb')
# Optional arguments
parser.add_argument('--chunk_size', type=int, default=1000, help='Chunk size (default: 1000)')
parser.add_argument('--chunk_overlap', type=int, default=200, help='Chunk overlap (default: 200)')
parser.add_argument(
'--db_type',
type=str,
default='faiss',
help='Type of vector store (default: faiss)',
)
args = parser.parse_args()
vectordb = VectorDb()
vectordb.create_vdb(
args.input_path,
args.output_db,
args.chunk_size,
args.chunk_overlap,
args.db_type,
)