Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import os
|
3 |
+
api_token = os.getenv("HF_TOKEN")
|
4 |
+
|
5 |
+
from langchain_community.vectorstores import FAISS
|
6 |
+
from langchain_community.document_loaders import PyPDFLoader
|
7 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
8 |
+
from langchain_community.embeddings import HuggingFaceEmbeddings
|
9 |
+
from langchain_community.llms import HuggingFaceEndpoint
|
10 |
+
from langchain.chains import ConversationalRetrievalChain
|
11 |
+
from langchain.memory import ConversationBufferMemory
|
12 |
+
|
13 |
+
list_llm = ["meta-llama/Llama-3.2-3B-Instruct", "mistralai/Mistral-7B-Instruct-v0.2"]
|
14 |
+
|
15 |
+
# Load and split PDF document
|
16 |
+
def load_doc(list_file_path):
|
17 |
+
loaders = [PyPDFLoader(x) for x in list_file_path]
|
18 |
+
pages = []
|
19 |
+
for loader in loaders:
|
20 |
+
pages.extend(loader.load())
|
21 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=64)
|
22 |
+
return text_splitter.split_documents(pages)
|
23 |
+
|
24 |
+
# Create vector database
|
25 |
+
def create_db(splits):
|
26 |
+
embeddings = HuggingFaceEmbeddings()
|
27 |
+
return FAISS.from_documents(splits, embeddings)
|
28 |
+
|
29 |
+
# Initialize langchain LLM chain
|
30 |
+
def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
|
31 |
+
llm = HuggingFaceEndpoint(
|
32 |
+
repo_id=llm_model,
|
33 |
+
huggingfacehub_api_token=api_token,
|
34 |
+
temperature=temperature,
|
35 |
+
max_new_tokens=max_tokens,
|
36 |
+
top_k=top_k,
|
37 |
+
)
|
38 |
+
|
39 |
+
memory = ConversationBufferMemory(memory_key="chat_history", output_key='answer', return_messages=True)
|
40 |
+
retriever = vector_db.as_retriever()
|
41 |
+
|
42 |
+
return ConversationalRetrievalChain.from_llm(
|
43 |
+
llm,
|
44 |
+
retriever=retriever,
|
45 |
+
chain_type="stuff",
|
46 |
+
memory=memory,
|
47 |
+
return_source_documents=True,
|
48 |
+
verbose=False,
|
49 |
+
)
|
50 |
+
|
51 |
+
st.title("RAG PDF Chatbot")
|
52 |
+
|
53 |
+
uploaded_files = st.file_uploader("Upload PDF files", accept_multiple_files=True, type="pdf")
|
54 |
+
|
55 |
+
if uploaded_files:
|
56 |
+
# Save uploaded files to local disk
|
57 |
+
file_paths = []
|
58 |
+
for uploaded_file in uploaded_files:
|
59 |
+
file_path = os.path.join("temp", uploaded_file.name)
|
60 |
+
os.makedirs("temp", exist_ok=True)
|
61 |
+
with open(file_path, "wb") as f:
|
62 |
+
f.write(uploaded_file.getbuffer())
|
63 |
+
file_paths.append(file_path)
|
64 |
+
|
65 |
+
st.session_state["doc_splits"] = load_doc(file_paths)
|
66 |
+
st.success("Documents successfully loaded and split!")
|
67 |
+
|
68 |
+
if 'vector_db' not in st.session_state and 'doc_splits' in st.session_state:
|
69 |
+
st.session_state['vector_db'] = create_db(st.session_state['doc_splits'])
|
70 |
+
|
71 |
+
llm_option = st.selectbox("Select LLM", list_llm)
|
72 |
+
|
73 |
+
temperature = st.slider("Temperature", 0.01, 1.0, 0.5, 0.1)
|
74 |
+
max_tokens = st.slider("Max Tokens", 128, 9192, 4096, 128)
|
75 |
+
top_k = st.slider("Top K", 1, 10, 3, 1)
|
76 |
+
|
77 |
+
if 'qa_chain' not in st.session_state and 'vector_db' in st.session_state:
|
78 |
+
st.session_state['qa_chain'] = initialize_llmchain(llm_option, temperature, max_tokens, top_k, st.session_state['vector_db'])
|
79 |
+
|
80 |
+
if "chat_history" not in st.session_state:
|
81 |
+
st.session_state["chat_history"] = []
|
82 |
+
|
83 |
+
user_input = st.text_input("Ask a question")
|
84 |
+
|
85 |
+
if st.button("Submit") and user_input:
|
86 |
+
qa_chain = st.session_state['qa_chain']
|
87 |
+
response = qa_chain.invoke({"question": user_input, "chat_history": st.session_state["chat_history"]})
|
88 |
+
|
89 |
+
st.session_state["chat_history"].append((user_input, response["answer"]))
|
90 |
+
|
91 |
+
st.write("### Response:")
|
92 |
+
st.write(response["answer"])
|
93 |
+
|
94 |
+
st.write("### Sources:")
|
95 |
+
for doc in response["source_documents"][:3]:
|
96 |
+
st.write(f"Page {doc.metadata['page'] + 1}: {doc.page_content[:300]}...")
|
97 |
+
|
98 |
+
st.write("### Chat History")
|
99 |
+
for user_msg, bot_msg in st.session_state["chat_history"]:
|
100 |
+
st.text(f"User: {user_msg}")
|
101 |
+
st.text(f"Assistant: {bot_msg}")
|