Create utils.py
Browse files
utils.py
ADDED
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from langchain_huggingface import HuggingFaceEmbeddings
|
2 |
+
from transformers import AutoModel, AutoTokenizer, AutoModelForCausalLM
|
3 |
+
from langchain_community.vectorstores import Chroma
|
4 |
+
from langchain.schema import Document
|
5 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
6 |
+
from langchain_community.llms.huggingface_pipeline import HuggingFacePipeline
|
7 |
+
import torch
|
8 |
+
|
9 |
+
embedding_model_name = 'nomic-ai/nomic-embed-text-v1.5'
|
10 |
+
|
11 |
+
model_kwargs = {'device':'cuda' if torch.cuda.is_available() else 'cpu',"trust_remote_code": True}
|
12 |
+
|
13 |
+
embeddings = HuggingFaceEmbeddings(
|
14 |
+
model_name=embedding_model_name,
|
15 |
+
model_kwargs=model_kwargs
|
16 |
+
)
|
17 |
+
|
18 |
+
vectorstore = None
|
19 |
+
|
20 |
+
|
21 |
+
|
22 |
+
def read_file(data: str) -> Document:
|
23 |
+
f = open(data,'r')
|
24 |
+
content = f.read()
|
25 |
+
f.close()
|
26 |
+
doc = Document(page_content=content, metadata={"name": data.split('/')[-1]})
|
27 |
+
return doc
|
28 |
+
|
29 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=800, chunk_overlap=100)
|
30 |
+
|
31 |
+
def add_doc(data,vectorstore):
|
32 |
+
doc = read_file(data)
|
33 |
+
splits = text_splitter.split_documents([doc])
|
34 |
+
vectorstore = Chroma.from_documents(documents=splits, embedding=embeddings)
|
35 |
+
retriever = vectorstore.as_retriever(search_kwargs={'k':1})
|
36 |
+
return retriever, vectorstore
|
37 |
+
|
38 |
+
def delete_doc(delete_name,vectorstore):
|
39 |
+
delete_doc_ids = []
|
40 |
+
for idx,name in enumerate(vectorstore.get()['metadatas']):
|
41 |
+
if name['name'] == delete_name:
|
42 |
+
delete_doc_ids.append(vectorstore.get()['ids'][idx])
|
43 |
+
for id in delete_doc_ids:
|
44 |
+
vectorstore.delete(ids = id)
|
45 |
+
# vectorstore.persist()
|
46 |
+
retriever = vectorstore.as_retriever(search_kwargs={'k':1})
|
47 |
+
return retriever, vectorstore
|
48 |
+
|
49 |
+
def delete_all_doc(vectorstore):
|
50 |
+
delete_doc_ids = vectorstore.get()['ids']
|
51 |
+
for id in delete_doc_ids:
|
52 |
+
vectorstore.delete(ids = id)
|
53 |
+
# vectorstore.persist()
|
54 |
+
retriever = vectorstore.as_retriever(search_kwargs={'k':1})
|
55 |
+
return retriever, vectorstore
|