Spaces:
Sleeping
Sleeping
create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,162 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from pypdf import PdfReader
|
3 |
+
import torch
|
4 |
+
from io import BytesIO
|
5 |
+
from langchain.prompts import PromptTemplate
|
6 |
+
from langchain.document_loaders import PyPDFLoader
|
7 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
8 |
+
from langchain.embeddings import HuggingFaceEmbeddings
|
9 |
+
from langchain.vectorstores import FAISS
|
10 |
+
from langchain.chains import RetrievalQA
|
11 |
+
import textwrap
|
12 |
+
|
13 |
+
from langchain.llms.huggingface_pipeline import HuggingFacePipeline
|
14 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, BitsAndBytesConfig
|
15 |
+
|
16 |
+
# load the environments
|
17 |
+
from dotenv import load_dotenv
|
18 |
+
load_dotenv()
|
19 |
+
|
20 |
+
|
21 |
+
|
22 |
+
#DEFINE SOME VARIABLES
|
23 |
+
|
24 |
+
CHUNK_SIZE = 1000
|
25 |
+
# Using HuggingFaceEmbeddings with the chosen embedding model
|
26 |
+
embeddings = HuggingFaceEmbeddings(
|
27 |
+
model_name="sentence-transformers/all-mpnet-base-v2",model_kwargs = {"device": "cuda"})
|
28 |
+
|
29 |
+
|
30 |
+
|
31 |
+
# transformer model configuration
|
32 |
+
quant_config = BitsAndBytesConfig(
|
33 |
+
load_in_4bit=True,
|
34 |
+
bnb_4bit_use_double_quant=True,
|
35 |
+
bnb_4bit_quant_type="nf4",
|
36 |
+
bnb_4bit_compute_dtype=torch.bfloat16
|
37 |
+
)
|
38 |
+
|
39 |
+
|
40 |
+
|
41 |
+
# CREATE A VECTOR DATABASE - FAISS
|
42 |
+
def creat_vector_db(uploaded_pdfs) -> FAISS:
|
43 |
+
"""Read multiple PDFs, split, embedd and store the embeddings on FAISS vector store"""
|
44 |
+
|
45 |
+
text = ""
|
46 |
+
for pdf in uploaded_pdfs:
|
47 |
+
pdf_reader = PdfReader(pdf)
|
48 |
+
for page in pdf_reader.pages:
|
49 |
+
text += page.extract_text()
|
50 |
+
|
51 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=CHUNK_SIZE,
|
52 |
+
chunk_overlap=100)
|
53 |
+
texts = text_splitter.split_text(text)
|
54 |
+
|
55 |
+
vector_db = FAISS.from_texts(texts, embeddings) # create vector db for similarity search
|
56 |
+
vector_db.save_local("faiss_index") # save the vector db to avoid repeated calls to it
|
57 |
+
return vector_db
|
58 |
+
|
59 |
+
# LOAD LLM
|
60 |
+
def load_llm():
|
61 |
+
|
62 |
+
model_id = "Deci/DeciLM-6b"
|
63 |
+
|
64 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
65 |
+
model = AutoModelForCausalLM.from_pretrained(model_id,
|
66 |
+
trust_remote_code=True,
|
67 |
+
device_map = "auto",
|
68 |
+
quantization_config=quant_config)
|
69 |
+
|
70 |
+
pipe = pipeline("text-generation",
|
71 |
+
model=model,
|
72 |
+
tokenizer=tokenizer,
|
73 |
+
temperature=0.1,
|
74 |
+
return_full_text = True,
|
75 |
+
max_new_tokens=40,
|
76 |
+
repetition_penalty = 1.1)
|
77 |
+
|
78 |
+
llm = HuggingFacePipeline(pipeline=pipe)
|
79 |
+
|
80 |
+
return llm
|
81 |
+
|
82 |
+
|
83 |
+
|
84 |
+
# RESPONSE INSTRUCTIONS
|
85 |
+
def set_custom_prompt():
|
86 |
+
"""instructions, to the llm for text response generation"""
|
87 |
+
|
88 |
+
|
89 |
+
custom_prompt_template = """You have been given the following documents to answer the user's question.
|
90 |
+
If you do not have information from the information given to answer the questions just say 'I don't know the answer" and don't try to make up an answer.
|
91 |
+
Context: {context}
|
92 |
+
Question: {question}
|
93 |
+
Give a detailed helpful answer and nothing more.
|
94 |
+
Helpful answer:
|
95 |
+
"""
|
96 |
+
prompt = PromptTemplate(template=custom_prompt_template, input_variables=[
|
97 |
+
"context", "question"])
|
98 |
+
return prompt
|
99 |
+
|
100 |
+
|
101 |
+
|
102 |
+
# QUESTION ANSWERING CHAIN
|
103 |
+
def retrieval_qa_chain(prompt, vector_db):
|
104 |
+
"""Chain to retrieve answers. the chain takes the documents and
|
105 |
+
makes a call to the DeciLM-6b llm """
|
106 |
+
|
107 |
+
llm = load_llm()
|
108 |
+
|
109 |
+
qa_chain = RetrievalQA.from_chain_type(
|
110 |
+
llm = llm,
|
111 |
+
chain_type = "stuff",
|
112 |
+
retriever = vector_db.as_retriever(),
|
113 |
+
return_source_documents=True,
|
114 |
+
chain_type_kwargs={"prompt": prompt}
|
115 |
+
)
|
116 |
+
|
117 |
+
return qa_chain
|
118 |
+
|
119 |
+
# QUESTION ANSWER BOT
|
120 |
+
def qa_bot():
|
121 |
+
vectore_db = FAISS.load_local("faiss_index", embeddings)
|
122 |
+
conversation_prompt = set_custom_prompt()
|
123 |
+
conversation = retrieval_qa_chain(conversation_prompt, vectore_db)
|
124 |
+
return conversation
|
125 |
+
|
126 |
+
# RESPONSE FROM BOT
|
127 |
+
def bot_response(query):
|
128 |
+
conversation_result = qa_bot()
|
129 |
+
response = conversation_result({"query": query})
|
130 |
+
return response["result"]
|
131 |
+
|
132 |
+
|
133 |
+
def main():
|
134 |
+
st.set_page_config(page_title="Multiple PDFs chat with DeciLM-6b and LangChain",
|
135 |
+
page_icon=":file_folder:")
|
136 |
+
|
137 |
+
# page side panel
|
138 |
+
with st.sidebar:
|
139 |
+
st.subheader("Hello, welcome!")
|
140 |
+
|
141 |
+
pdfs = st.file_uploader(label="Upload your PDFs here and click Process!",
|
142 |
+
accept_multiple_files=True)
|
143 |
+
|
144 |
+
if st.button("Process"):
|
145 |
+
with st.spinner("Processing file(s)..."):
|
146 |
+
# create a vectore store
|
147 |
+
creat_vector_db(pdfs)
|
148 |
+
st.write("Your files are Processed. You set to ask questions!")
|
149 |
+
|
150 |
+
st.header("Chat with Multiple PDFs using DeciLM-6b-instruct LLM")
|
151 |
+
|
152 |
+
# Query side
|
153 |
+
query = st.text_input(label="Type your question based on the PDFs",
|
154 |
+
placeholder="Type question...")
|
155 |
+
|
156 |
+
if query:
|
157 |
+
st.write(f"Query: {query}")
|
158 |
+
st.text(textwrap.fill(bot_response(query), width=80))
|
159 |
+
|
160 |
+
|
161 |
+
if __name__ == "__main__":
|
162 |
+
main()
|