Spaces:
Sleeping
Sleeping
Add new app icon. Refactor and cleanup.
Browse files- README.md +1 -1
- app.py +52 -107
- assets/app_icon.png +0 -0
- assets/icon.jpg +0 -0
- assets/large_icon.png +0 -0
- document_retriever.py +58 -0
README.md
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
<div align="center">
|
2 |
-
<img alt="app icon" height="196px" src="./assets/
|
3 |
</div>
|
4 |
|
5 |
<div align="center">
|
|
|
1 |
<div align="center">
|
2 |
+
<img alt="app icon" height="196px" src="./assets/app_icon.jpg">
|
3 |
</div>
|
4 |
|
5 |
<div align="center">
|
app.py
CHANGED
@@ -1,27 +1,16 @@
|
|
1 |
-
import os
|
2 |
-
import tempfile
|
3 |
-
|
4 |
import streamlit as st
|
|
|
5 |
from langchain.chains import ConversationalRetrievalChain
|
6 |
from langchain.chat_models import ChatOpenAI
|
7 |
-
from langchain.embeddings import HuggingFaceEmbeddings
|
8 |
from langchain.memory import ConversationBufferMemory
|
9 |
from langchain.memory.chat_message_histories import StreamlitChatMessageHistory
|
10 |
-
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
11 |
-
from langchain_community.document_loaders import (
|
12 |
-
Docx2txtLoader,
|
13 |
-
PyPDFLoader,
|
14 |
-
TextLoader,
|
15 |
-
UnstructuredEPubLoader,
|
16 |
-
)
|
17 |
-
from langchain_community.vectorstores import DocArrayInMemorySearch
|
18 |
|
|
|
19 |
from calback_handler import PrintRetrievalHandler, StreamHandler
|
20 |
from chat_profile import ChatProfileRoleEnum
|
21 |
|
22 |
# configs
|
23 |
LLM_MODEL_NAME = "gpt-3.5-turbo"
|
24 |
-
EMBEDDING_MODEL_NAME = "all-MiniLM-L6-v2"
|
25 |
|
26 |
st.set_page_config(
|
27 |
page_title=":books: InkChatGPT: Chat with Documents",
|
@@ -34,115 +23,71 @@ st.set_page_config(
|
|
34 |
},
|
35 |
)
|
36 |
|
37 |
-
st.image("./assets/icon.jpg", width=100)
|
38 |
-
st.header(
|
39 |
-
":gray[:books: InkChatGPT]",
|
40 |
-
divider="blue",
|
41 |
-
)
|
42 |
-
st.write("**Chat** with Documents")
|
43 |
-
|
44 |
# Setup memory for contextual conversation
|
45 |
msgs = StreamlitChatMessageHistory()
|
46 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
|
48 |
-
|
49 |
-
def configure_retriever(files):
|
50 |
-
# Read documents
|
51 |
-
docs = []
|
52 |
-
temp_dir = tempfile.TemporaryDirectory()
|
53 |
-
for file in files:
|
54 |
-
temp_filepath = os.path.join(temp_dir.name, file.name)
|
55 |
-
with open(temp_filepath, "wb") as f:
|
56 |
-
f.write(file.getvalue())
|
57 |
-
|
58 |
-
_, extension = os.path.splitext(temp_filepath)
|
59 |
-
|
60 |
-
# Load the file using the appropriate loader
|
61 |
-
if extension == ".pdf":
|
62 |
-
loader = PyPDFLoader(temp_filepath)
|
63 |
-
elif extension == ".docx":
|
64 |
-
loader = Docx2txtLoader(temp_filepath)
|
65 |
-
elif extension == ".txt":
|
66 |
-
loader = TextLoader(temp_filepath)
|
67 |
-
elif extension == ".epub":
|
68 |
-
loader = UnstructuredEPubLoader(temp_filepath)
|
69 |
-
else:
|
70 |
-
st.write("This document format is not supported!")
|
71 |
-
return None
|
72 |
-
|
73 |
-
# loader = PyPDFLoader(temp_filepath)
|
74 |
-
docs.extend(loader.load())
|
75 |
-
|
76 |
-
# Split documents
|
77 |
-
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1500, chunk_overlap=200)
|
78 |
-
splits = text_splitter.split_documents(docs)
|
79 |
-
|
80 |
-
# Create embeddings and store in vectordb
|
81 |
-
embeddings = HuggingFaceEmbeddings(model_name=EMBEDDING_MODEL_NAME)
|
82 |
-
vectordb = DocArrayInMemorySearch.from_documents(splits, embeddings)
|
83 |
-
|
84 |
-
# Define retriever
|
85 |
-
retriever = vectordb.as_retriever(
|
86 |
-
search_type="mmr", search_kwargs={"k": 2, "fetch_k": 4}
|
87 |
-
)
|
88 |
-
|
89 |
-
return retriever
|
90 |
-
|
91 |
-
|
92 |
-
with st.sidebar.expander("Documents"):
|
93 |
-
st.subheader("Files")
|
94 |
uploaded_files = st.file_uploader(
|
95 |
label="Select files",
|
96 |
type=["pdf", "txt", "docx", "epub"],
|
97 |
accept_multiple_files=True,
|
98 |
)
|
99 |
|
100 |
-
with
|
101 |
-
|
102 |
-
|
103 |
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
msgs.add_ai_message("How can I help you?")
|
108 |
|
109 |
-
|
110 |
-
|
111 |
-
|
|
|
|
|
|
|
|
|
112 |
|
113 |
-
|
114 |
-
|
|
|
115 |
|
116 |
-
|
117 |
-
|
118 |
-
|
|
|
119 |
|
120 |
-
|
121 |
-
|
122 |
-
model_name=LLM_MODEL_NAME,
|
123 |
-
openai_api_key=openai_api_key,
|
124 |
-
temperature=0,
|
125 |
-
streaming=True,
|
126 |
-
)
|
127 |
-
|
128 |
-
chain = ConversationalRetrievalChain.from_llm(
|
129 |
-
llm, retriever=result_retriever, memory=memory, verbose=False
|
130 |
-
)
|
131 |
|
132 |
-
|
133 |
-
|
134 |
-
ChatProfileRoleEnum.AI: "assistant",
|
135 |
-
}
|
136 |
-
|
137 |
-
for msg in msgs.messages:
|
138 |
-
st.chat_message(avatars[msg.type]).write(msg.content)
|
139 |
|
140 |
-
|
141 |
-
|
|
|
|
|
142 |
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
user_query, callbacks=[retrieval_handler, stream_handler]
|
148 |
-
)
|
|
|
|
|
|
|
|
|
1 |
import streamlit as st
|
2 |
+
|
3 |
from langchain.chains import ConversationalRetrievalChain
|
4 |
from langchain.chat_models import ChatOpenAI
|
|
|
5 |
from langchain.memory import ConversationBufferMemory
|
6 |
from langchain.memory.chat_message_histories import StreamlitChatMessageHistory
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
|
8 |
+
from document_retriever import configure_retriever
|
9 |
from calback_handler import PrintRetrievalHandler, StreamHandler
|
10 |
from chat_profile import ChatProfileRoleEnum
|
11 |
|
12 |
# configs
|
13 |
LLM_MODEL_NAME = "gpt-3.5-turbo"
|
|
|
14 |
|
15 |
st.set_page_config(
|
16 |
page_title=":books: InkChatGPT: Chat with Documents",
|
|
|
23 |
},
|
24 |
)
|
25 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
# Setup memory for contextual conversation
|
27 |
msgs = StreamlitChatMessageHistory()
|
28 |
|
29 |
+
with st.container():
|
30 |
+
col1, col2 = st.columns([0.2, 0.8])
|
31 |
+
with col1:
|
32 |
+
st.image(
|
33 |
+
"./assets/large_icon.png", use_column_width="always", output_format="PNG"
|
34 |
+
)
|
35 |
+
with col2:
|
36 |
+
st.header(":books: InkChatGPT")
|
37 |
+
st.write("**Chat** with Documents")
|
38 |
+
st.caption("Supports PDF, TXT, DOCX, EPUB β’ Limit 200MB per file")
|
39 |
+
|
40 |
+
chat_tab, documents_tab, settings_tab = st.tabs(["Chat", "Documents", "Settings"])
|
41 |
+
with settings_tab:
|
42 |
+
openai_api_key = st.text_input("OpenAI API Key", type="password")
|
43 |
+
if len(msgs.messages) == 0 or st.button("Clear message history"):
|
44 |
+
msgs.clear()
|
45 |
+
msgs.add_ai_message("How can I help you?")
|
46 |
|
47 |
+
with documents_tab:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
48 |
uploaded_files = st.file_uploader(
|
49 |
label="Select files",
|
50 |
type=["pdf", "txt", "docx", "epub"],
|
51 |
accept_multiple_files=True,
|
52 |
)
|
53 |
|
54 |
+
with chat_tab:
|
55 |
+
if uploaded_files:
|
56 |
+
result_retriever = configure_retriever(uploaded_files)
|
57 |
|
58 |
+
memory = ConversationBufferMemory(
|
59 |
+
memory_key="chat_history", chat_memory=msgs, return_messages=True
|
60 |
+
)
|
|
|
61 |
|
62 |
+
# Setup LLM and QA chain
|
63 |
+
llm = ChatOpenAI(
|
64 |
+
model_name=LLM_MODEL_NAME,
|
65 |
+
openai_api_key=openai_api_key,
|
66 |
+
temperature=0,
|
67 |
+
streaming=True,
|
68 |
+
)
|
69 |
|
70 |
+
chain = ConversationalRetrievalChain.from_llm(
|
71 |
+
llm, retriever=result_retriever, memory=memory, verbose=False
|
72 |
+
)
|
73 |
|
74 |
+
avatars = {
|
75 |
+
ChatProfileRoleEnum.Human: "user",
|
76 |
+
ChatProfileRoleEnum.AI: "assistant",
|
77 |
+
}
|
78 |
|
79 |
+
for msg in msgs.messages:
|
80 |
+
st.chat_message(avatars[msg.type]).write(msg.content)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
81 |
|
82 |
+
if not openai_api_key:
|
83 |
+
st.caption("π Add your **OpenAI API key** on the `Settings` to continue.")
|
|
|
|
|
|
|
|
|
|
|
84 |
|
85 |
+
if user_query := st.chat_input(
|
86 |
+
placeholder="Ask me anything!", disabled=(not openai_api_key)
|
87 |
+
):
|
88 |
+
st.chat_message("user").write(user_query)
|
89 |
|
90 |
+
with st.chat_message("assistant"):
|
91 |
+
retrieval_handler = PrintRetrievalHandler(st.empty())
|
92 |
+
stream_handler = StreamHandler(st.empty())
|
93 |
+
response = chain.run(user_query, callbacks=[retrieval_handler, stream_handler])
|
|
|
|
assets/app_icon.png
ADDED
![]() |
assets/icon.jpg
DELETED
Binary file (49.5 kB)
|
|
assets/large_icon.png
ADDED
![]() |
document_retriever.py
ADDED
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import tempfile
|
3 |
+
|
4 |
+
import streamlit as st
|
5 |
+
from langchain.embeddings import HuggingFaceEmbeddings
|
6 |
+
from langchain_community.document_loaders import (
|
7 |
+
Docx2txtLoader,
|
8 |
+
PyPDFLoader,
|
9 |
+
TextLoader,
|
10 |
+
UnstructuredEPubLoader,
|
11 |
+
)
|
12 |
+
from langchain_community.vectorstores import DocArrayInMemorySearch
|
13 |
+
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
14 |
+
|
15 |
+
EMBEDDING_MODEL_NAME = "all-MiniLM-L6-v2"
|
16 |
+
|
17 |
+
|
18 |
+
@st.cache_resource(ttl="1h")
|
19 |
+
def configure_retriever(files):
|
20 |
+
# Read documents
|
21 |
+
docs = []
|
22 |
+
temp_dir = tempfile.TemporaryDirectory()
|
23 |
+
for file in files:
|
24 |
+
temp_filepath = os.path.join(temp_dir.name, file.name)
|
25 |
+
with open(temp_filepath, "wb") as f:
|
26 |
+
f.write(file.getvalue())
|
27 |
+
|
28 |
+
_, extension = os.path.splitext(temp_filepath)
|
29 |
+
|
30 |
+
# Load the file using the appropriate loader
|
31 |
+
if extension == ".pdf":
|
32 |
+
loader = PyPDFLoader(temp_filepath)
|
33 |
+
elif extension == ".docx":
|
34 |
+
loader = Docx2txtLoader(temp_filepath)
|
35 |
+
elif extension == ".txt":
|
36 |
+
loader = TextLoader(temp_filepath)
|
37 |
+
elif extension == ".epub":
|
38 |
+
loader = UnstructuredEPubLoader(temp_filepath)
|
39 |
+
else:
|
40 |
+
st.write("This document format is not supported!")
|
41 |
+
return None
|
42 |
+
|
43 |
+
docs.extend(loader.load())
|
44 |
+
|
45 |
+
# Split documents
|
46 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1500, chunk_overlap=200)
|
47 |
+
splits = text_splitter.split_documents(docs)
|
48 |
+
|
49 |
+
# Create embeddings and store in vectordb
|
50 |
+
embeddings = HuggingFaceEmbeddings(model_name=EMBEDDING_MODEL_NAME)
|
51 |
+
vectordb = DocArrayInMemorySearch.from_documents(splits, embeddings)
|
52 |
+
|
53 |
+
# Define retriever
|
54 |
+
retriever = vectordb.as_retriever(
|
55 |
+
search_type="mmr", search_kwargs={"k": 2, "fetch_k": 4}
|
56 |
+
)
|
57 |
+
|
58 |
+
return retriever
|