Update app.py
Browse files
app.py
CHANGED
@@ -1,145 +1,84 @@
|
|
|
|
|
|
|
|
1 |
import streamlit as st
|
2 |
-
from
|
3 |
-
from
|
4 |
-
from
|
5 |
-
from langchain.embeddings import OpenAIEmbeddings, HuggingFaceInstructEmbeddings, HuggingFaceEmbeddings
|
6 |
-
from langchain.vectorstores import FAISS
|
7 |
-
from langchain.chat_models import ChatOpenAI
|
8 |
from langchain.memory import ConversationBufferMemory
|
9 |
from langchain.chains import ConversationalRetrievalChain
|
10 |
-
from langchain.chat_models.gigachat import GigaChat
|
11 |
-
from htmlTemplates import css, bot_template, user_template
|
12 |
-
from langchain.llms import HuggingFaceHub, LlamaCpp
|
13 |
-
from huggingface_hub import snapshot_download, hf_hub_download
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
repo_name = "IlyaGusev/saiga_mistral_7b_gguf"
|
18 |
-
model_name = "model-q4_K.gguf"
|
19 |
-
|
20 |
-
#snapshot_download(repo_id=repo_name, local_dir=".", allow_patterns=model_name)
|
21 |
-
|
22 |
-
from transformers import pipeline
|
23 |
-
|
24 |
-
# Initialize the summarization pipeline
|
25 |
-
summarizer = pipeline("summarization")
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
def get_pdf_text(pdf_docs):
|
30 |
-
text = ""
|
31 |
-
for pdf in pdf_docs:
|
32 |
-
pdf_reader = PdfReader(pdf)
|
33 |
-
for page in pdf_reader.pages:
|
34 |
-
text += page.extract_text()
|
35 |
-
|
36 |
-
return text
|
37 |
-
|
38 |
-
|
39 |
-
def get_text_chunks(text):
|
40 |
-
text_splitter = CharacterTextSplitter(separator="\n",
|
41 |
-
chunk_size=1000, # 1000
|
42 |
-
chunk_overlap=200, # 200
|
43 |
-
length_function=len
|
44 |
-
)
|
45 |
-
chunks = text_splitter.split_text(text)
|
46 |
-
|
47 |
-
return chunks
|
48 |
|
|
|
49 |
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
#embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/paraphrase-multilingual-mpnet-base-v2")
|
55 |
-
#vectorstore = FAISS.from_texts(texts=text_chunks, embedding=embeddings)
|
56 |
|
57 |
-
#return vectorstore
|
58 |
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
vectorstore =
|
|
|
63 |
return vectorstore
|
64 |
|
65 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
llm = GigaChat(profanity=False,
|
71 |
-
verify_ssl_certs=False
|
72 |
-
)
|
73 |
-
|
74 |
-
memory = ConversationBufferMemory(memory_key='chat_history',
|
75 |
-
input_key='question',
|
76 |
-
output_key='answer',
|
77 |
-
return_messages=True)
|
78 |
-
|
79 |
-
conversation_chain = ConversationalRetrievalChain.from_llm(llm=llm,
|
80 |
-
retriever=vectorstore.as_retriever(),
|
81 |
-
memory=memory,
|
82 |
-
return_source_documents=True
|
83 |
-
)
|
84 |
-
|
85 |
-
return conversation_chain
|
86 |
-
|
87 |
-
|
88 |
-
def summarize_text(text):
|
89 |
-
summary = summarizer(text, max_length=130, min_length=30, do_sample=False)
|
90 |
-
return summary[0]['summary_text']
|
91 |
-
|
92 |
-
|
93 |
-
def handle_userinput(user_question):
|
94 |
-
response = st.session_state.conversation({'question': user_question})
|
95 |
-
|
96 |
-
st.session_state.chat_history = response['chat_history']
|
97 |
-
st.session_state.retrieved_text = response['source_documents']
|
98 |
-
|
99 |
-
for i, (message, text) in enumerate(zip(st.session_state.chat_history, st.session_state.retrieved_text)):
|
100 |
-
if i % 2 == 0: # User messages
|
101 |
-
st.write(user_template.replace("{{MSG}}", message.content), unsafe_allow_html=True)
|
102 |
-
else: # Bot messages
|
103 |
-
st.write(bot_template.replace("{{MSG}}", message.content), unsafe_allow_html=True)
|
104 |
-
if summarize_option and text.page_content: # Check if summarization is enabled
|
105 |
-
summarized_text = summarize_text(text.page_content)
|
106 |
-
st.write(bot_template.replace("{{MSG}}", summarized_text), unsafe_allow_html=True)
|
107 |
-
else:
|
108 |
-
st.write(bot_template.replace("{{MSG}}", text.page_content), unsafe_allow_html=True)
|
109 |
-
|
110 |
|
|
|
|
|
111 |
|
112 |
-
|
113 |
-
|
114 |
-
st.write(css, unsafe_allow_html=True)
|
115 |
|
116 |
-
if "conversation" not in st.session_state:
|
117 |
-
st.session_state.conversation = None
|
118 |
-
if "chat_history" not in st.session_state:
|
119 |
-
st.session_state.chat_history = None
|
120 |
|
121 |
-
|
122 |
-
|
|
|
123 |
|
124 |
-
|
125 |
-
handle_userinput(user_question)
|
126 |
|
127 |
-
|
128 |
-
st.
|
129 |
-
embedding_model_name = st.selectbox("Select embedding model", ["intfloat/multilingual-e5-large", "sentence-transformers/paraphrase-multilingual-mpnet-base-v2"])
|
130 |
-
summarize_option = st.sidebar.checkbox("Enable Summarization", value=False)
|
131 |
-
pdf_docs = st.file_uploader(
|
132 |
-
"Upload your PDFs here and click on 'Process'", accept_multiple_files=True)
|
133 |
-
if st.button("Process"):
|
134 |
-
with st.spinner("Processing"):
|
135 |
-
# get pdf text
|
136 |
-
raw_text = get_pdf_text(pdf_docs)
|
137 |
|
138 |
-
|
139 |
-
|
140 |
|
141 |
-
# create vector store
|
142 |
-
vectorstore = get_vectorstore(text_chunks, embedding_model_name)
|
143 |
|
144 |
-
|
145 |
-
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import json
|
3 |
+
|
4 |
import streamlit as st
|
5 |
+
from langchain_huggingface import HuggingFaceEmbeddings
|
6 |
+
from langchain_chroma import Chroma
|
7 |
+
from langchain_groq import ChatGroq
|
|
|
|
|
|
|
8 |
from langchain.memory import ConversationBufferMemory
|
9 |
from langchain.chains import ConversationalRetrievalChain
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
|
11 |
+
from vectorize_documents import embeddings
|
12 |
|
13 |
+
working_dir = os.path.dirname(os.path.abspath(__file__))
|
14 |
+
config_data = json.load(open(f"{working_dir}/config.json"))
|
15 |
+
GROQ_API_KEY = config_data["GROQ_API_KEY"]
|
16 |
+
os.environ["GROQ_API_KEY"] = GROQ_API_KEY
|
|
|
|
|
17 |
|
|
|
18 |
|
19 |
+
def setup_vectorstore():
|
20 |
+
persist_directory = f"{working_dir}/vector_db_dir"
|
21 |
+
embedddings = HuggingFaceEmbeddings()
|
22 |
+
vectorstore = Chroma(persist_directory=persist_directory,
|
23 |
+
embedding_function=embeddings)
|
24 |
return vectorstore
|
25 |
|
26 |
|
27 |
+
def chat_chain(vectorstore):
|
28 |
+
llm = ChatGroq(model="llama-3.1-70b-versatile",
|
29 |
+
temperature=0)
|
30 |
+
retriever = vectorstore.as_retriever()
|
31 |
+
memory = ConversationBufferMemory(
|
32 |
+
llm=llm,
|
33 |
+
output_key="answer",
|
34 |
+
memory_key="chat_history",
|
35 |
+
return_messages=True
|
36 |
+
)
|
37 |
+
chain = ConversationalRetrievalChain.from_llm(
|
38 |
+
llm=llm,
|
39 |
+
retriever=retriever,
|
40 |
+
chain_type="stuff",
|
41 |
+
memory=memory,
|
42 |
+
verbose=True,
|
43 |
+
return_source_documents=True
|
44 |
+
)
|
45 |
+
|
46 |
+
return chain
|
47 |
+
|
48 |
+
|
49 |
+
st.set_page_config(
|
50 |
+
page_title="Multi Doc Chat",
|
51 |
+
page_icon = "π",
|
52 |
+
layout="centered"
|
53 |
+
)
|
54 |
+
|
55 |
+
st.title("π Multi Documents Chatbot")
|
56 |
|
57 |
+
if "chat_history" not in st.session_state:
|
58 |
+
st.session_state.chat_history = []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
|
60 |
+
if "vectorstore" not in st.session_state:
|
61 |
+
st.session_state.vectorstore = setup_vectorstore()
|
62 |
|
63 |
+
if "conversationsal_chain" not in st.session_state:
|
64 |
+
st.session_state.conversationsal_chain = chat_chain(st.session_state.vectorstore)
|
|
|
65 |
|
|
|
|
|
|
|
|
|
66 |
|
67 |
+
for message in st.session_state.chat_history:
|
68 |
+
with st.chat_message(message["role"]):
|
69 |
+
st.markdown(message["content"])
|
70 |
|
71 |
+
user_input = st.chat_input("Ask AI...")
|
|
|
72 |
|
73 |
+
if user_input:
|
74 |
+
st.session_state.chat_history.append({"role": "user", "content": user_input})
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
|
76 |
+
with st.chat_message("user"):
|
77 |
+
st.markdown(user_input)
|
78 |
|
|
|
|
|
79 |
|
80 |
+
with st.chat_message("assistant"):
|
81 |
+
response = st.session_state.conversationsal_chain({"question": user_input})
|
82 |
+
assistant_response = response["answer"]
|
83 |
+
st.markdown(assistant_response)
|
84 |
+
st.session_state.chat_history.append({"role": "assistant", "content": assistant_response})
|