-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
70 lines (54 loc) · 1.98 KB
/
utils.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
from langchain.memory import ConversationBufferWindowMemory, ChatMessageHistory
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.document_loaders import PyPDFLoader
from langchain_community.embeddings import OllamaEmbeddings
from langchain_community.vectorstores import FAISS
from config import settings
import os
def load_documents():
"""
Loads and splits documents from a PDF file.
Returns:
list: A list of document chunks.
"""
loader = PyPDFLoader(settings.FILE_PATH)
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
return loader.load_and_split(text_splitter)
def get_embeddings():
"""
Initializes and returns the embeddings model.
Returns:
OllamaEmbeddings: The embeddings model.
"""
return OllamaEmbeddings(model=settings.EMBEDDINGS_MODEL)
def get_memory(model):
"""
Initializes and returns the memory buffer and chat history.
Args:
model (ChatOpenAI): The chat model.
Returns:
tuple: A tuple containing the memory buffer and chat history.
"""
history = ChatMessageHistory()
memory_buffer = ConversationBufferWindowMemory(
llm=model,
return_messages=True,
memory_key="chat_history",
chat_memory=history
)
return memory_buffer, history
def get_retriever(embeddings, documents):
"""
Initializes and returns the retriever.
Args:
embeddings (OllamaEmbeddings): The embeddings model.
documents (list): The list of document chunks.
Returns:
FAISS: The retriever object.
"""
if os.path.exists("vectorstore"):
faiss_index = FAISS.load_local("vectorstore", embeddings, allow_dangerous_deserialization=True)
else:
faiss_index = FAISS.from_documents(documents=documents, embedding=embeddings)
faiss_index.save_local("vectorstore")
return faiss_index.as_retriever(search_type="similarity", search_kwargs={"k": 3})