Spaces:
Sleeping
Sleeping
Add Streamlit application and dependencies
Browse files- app.py +66 -0
- requirements.txt +7 -0
app.py
ADDED
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from PyPDF2 import PdfReader
|
3 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
4 |
+
from transformers import pipeline, AutoTokenizer, AutoModelForQuestionAnswering
|
5 |
+
from sentence_transformers import SentenceTransformer
|
6 |
+
import faiss
|
7 |
+
import numpy as np
|
8 |
+
|
9 |
+
# Load the lightweight Hugging Face transformer model
|
10 |
+
model_name = "distilbert-base-uncased-distilled-squad"
|
11 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
12 |
+
qa_model = AutoModelForQuestionAnswering.from_pretrained(model_name)
|
13 |
+
qa_pipeline = pipeline("question-answering", model=qa_model, tokenizer=tokenizer)
|
14 |
+
|
15 |
+
# Load the SentenceTransformer model for embeddings
|
16 |
+
embedding_model = SentenceTransformer('all-MiniLM-L6-v2')
|
17 |
+
|
18 |
+
# Upload PDF files
|
19 |
+
st.header("Question and Answer Chatbot")
|
20 |
+
|
21 |
+
with st.sidebar:
|
22 |
+
st.title("Turn your PDFs into a Q&A session. Upload a file and start asking questions")
|
23 |
+
file = st.file_uploader("PDF file upload", type="pdf")
|
24 |
+
|
25 |
+
# Extract the text
|
26 |
+
if file is not None:
|
27 |
+
pdf_reader = PdfReader(file)
|
28 |
+
text = ""
|
29 |
+
for page in pdf_reader.pages:
|
30 |
+
text += page.extract_text()
|
31 |
+
|
32 |
+
# Break it into chunks
|
33 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
34 |
+
separators="\n",
|
35 |
+
#chunk_size=1000,
|
36 |
+
#chunk_overlap=500,
|
37 |
+
chunk_size=800,
|
38 |
+
chunk_overlap=150,
|
39 |
+
length_function=len
|
40 |
+
)
|
41 |
+
chunks = text_splitter.split_text(text)
|
42 |
+
|
43 |
+
# Generate embeddings for each chunk
|
44 |
+
embeddings = embedding_model.encode(chunks)
|
45 |
+
|
46 |
+
# Create FAISS index and add embeddings
|
47 |
+
dimension = embeddings.shape[1]
|
48 |
+
index = faiss.IndexFlatL2(dimension)
|
49 |
+
index.add(np.array(embeddings))
|
50 |
+
|
51 |
+
# Get user question
|
52 |
+
user_question = st.text_input("Type your question here")
|
53 |
+
|
54 |
+
# Perform similarity search
|
55 |
+
if user_question:
|
56 |
+
question_embedding = embedding_model.encode([user_question])
|
57 |
+
D, I = index.search(np.array(question_embedding), k=5)
|
58 |
+
matched_texts = [chunks[i] for i in I[0]]
|
59 |
+
|
60 |
+
# Use the lightweight transformer model for question answering
|
61 |
+
response = ""
|
62 |
+
for context in matched_texts:
|
63 |
+
result = qa_pipeline(question=user_question, context=context)
|
64 |
+
response += result['answer'] + " "
|
65 |
+
|
66 |
+
st.write(response)
|
requirements.txt
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
streamlit
|
2 |
+
PyPDF2
|
3 |
+
langchain # Ensure this package name is correct based on your actual usage
|
4 |
+
transformers
|
5 |
+
sentence-transformers
|
6 |
+
faiss
|
7 |
+
numpy
|