app.py

import streamlit as st
import tempfile
import torch
from langchain.document_loaders import PyPDFLoader

from langchain.text_splitter import SentenceTransformersTokenTextSplitter
from langchain.embeddings import HuggingFaceEmbeddings, SentenceTransformerEmbeddings
from langchain.vectorstores import FAISS


model_name = "dangvantuan/sentence-camembert-large"
model_kwargs = {'device': 'cpu'}
encode_kwargs = {'convert_to_tensor': True}
splitter = SentenceTransformersTokenTextSplitter(model_name=model_name,
                                                 tokens_per_chunk=380,
                                                 chunk_overlap=100)

embeddings_fun = HuggingFaceEmbeddings(
    model_name=model_name,
    model_kwargs=model_kwargs,
    encode_kwargs=encode_kwargs
)


def read_pdf(file):
    with tempfile.NamedTemporaryFile(delete=False) as temp:
        temp.write(file.getvalue())
        loader = PyPDFLoader(temp.name)
        raw_documents = loader.load()
        return raw_documents

st.title('PDF Text Extractor')

uploaded_file = st.file_uploader("Choose a PDF file", type="pdf")

query = st.text_input("Entrer une question")

st.text('La reponse à votre question:')

if uploaded_file is not None:
    raw_documents = read_pdf(uploaded_file)    
    documents = splitter.split_documents(raw_documents)
    
    # embeddings_text = embeddings_fun.embed_documents(documents)
    
    faiss_db = FAISS.from_documents(documents, embeddings_fun)
    
    docs = faiss_db.similarity_search(query)
    
    st.write(docs[0].page_content)
else:
    st.write("file not uploaded correctly")