Spaces:
Sleeping
Sleeping
surprisedPikachu007
commited on
Commit
•
f13531e
1
Parent(s):
5c30eba
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,134 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from langchain_community.document_loaders import PyPDFLoader
|
3 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
4 |
+
from langchain_community.vectorstores import Chroma
|
5 |
+
from langchain.chains import ConversationalRetrievalChain
|
6 |
+
from langchain_community.embeddings import HuggingFaceEmbeddings
|
7 |
+
from langchain_community.llms import HuggingFacePipeline, HuggingFaceEndpoint
|
8 |
+
from langchain.memory import ConversationBufferMemory
|
9 |
+
from pathlib import Path
|
10 |
+
import chromadb
|
11 |
+
import re
|
12 |
+
|
13 |
+
def load_doc(list_file_path, chunk_size=600, chunk_overlap=40):
|
14 |
+
loaders = [PyPDFLoader(x) for x in list_file_path]
|
15 |
+
pages = [page for loader in loaders for page in loader.load()]
|
16 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)
|
17 |
+
doc_splits = text_splitter.split_documents(pages)
|
18 |
+
return doc_splits
|
19 |
+
|
20 |
+
def create_db(splits, collection_name):
|
21 |
+
embedding = HuggingFaceEmbeddings()
|
22 |
+
client = chromadb.EphemeralClient()
|
23 |
+
vectordb = Chroma.from_documents(
|
24 |
+
documents=splits,
|
25 |
+
embedding=embedding,
|
26 |
+
client=client,
|
27 |
+
collection_name=collection_name,
|
28 |
+
)
|
29 |
+
return vectordb
|
30 |
+
|
31 |
+
def initialize_llmchain(llm_model, vector_db, progress=gr.Progress()):
|
32 |
+
llm = HuggingFaceEndpoint(
|
33 |
+
repo_id=llm_model,
|
34 |
+
temperature=0.7,
|
35 |
+
max_new_tokens=1024,
|
36 |
+
top_k=3,
|
37 |
+
)
|
38 |
+
memory = ConversationBufferMemory(memory_key="chat_history", output_key='answer', return_messages=True)
|
39 |
+
retriever = vector_db.as_retriever()
|
40 |
+
qa_chain = ConversationalRetrievalChain.from_llm(
|
41 |
+
llm,
|
42 |
+
retriever=retriever,
|
43 |
+
chain_type="stuff",
|
44 |
+
memory=memory,
|
45 |
+
return_source_documents=True,
|
46 |
+
verbose=False,
|
47 |
+
)
|
48 |
+
return qa_chain
|
49 |
+
|
50 |
+
def create_collection_name(filepath):
|
51 |
+
collection_name = Path(filepath).stem
|
52 |
+
collection_name = re.sub('[^A-Za-z0-9]+', '-', collection_name)[:50]
|
53 |
+
if len(collection_name) < 3:
|
54 |
+
collection_name += 'xyz'
|
55 |
+
if not collection_name[0].isalnum():
|
56 |
+
collection_name = 'A' + collection_name[1:]
|
57 |
+
if not collection_name[-1].isalnum():
|
58 |
+
collection_name = collection_name[:-1] + 'Z'
|
59 |
+
return collection_name
|
60 |
+
|
61 |
+
def initialize_database(list_file_obj, progress=gr.Progress()):
|
62 |
+
list_file_path = [x.name for x in list_file_obj if x is not None]
|
63 |
+
collection_name = create_collection_name(list_file_path[0])
|
64 |
+
doc_splits = load_doc(list_file_path)
|
65 |
+
vector_db = create_db(doc_splits, collection_name)
|
66 |
+
return vector_db, collection_name, "Complete!"
|
67 |
+
|
68 |
+
def initialize_LLM(llm_model, vector_db, progress=gr.Progress()):
|
69 |
+
qa_chain = initialize_llmchain(llm_model, vector_db, progress)
|
70 |
+
return qa_chain, "Complete!"
|
71 |
+
|
72 |
+
def conversation(qa_chain, message, history):
|
73 |
+
formatted_chat_history = [(f"User: {user_message}", f"Assistant: {bot_message}") for user_message, bot_message in history]
|
74 |
+
response = qa_chain({"question": message, "chat_history": formatted_chat_history})
|
75 |
+
response_answer = response["answer"]
|
76 |
+
if "Helpful Answer:" in response_answer:
|
77 |
+
response_answer = response_answer.split("Helpful Answer:")[-1]
|
78 |
+
response_sources = response["source_documents"]
|
79 |
+
response_source1 = response_sources[0].page_content.strip()
|
80 |
+
response_source2 = response_sources[1].page_content.strip()
|
81 |
+
response_source3 = response_sources[2].page_content.strip()
|
82 |
+
response_source1_page = response_sources[0].metadata["page"] + 1
|
83 |
+
response_source2_page = response_sources[1].metadata["page"] + 1
|
84 |
+
response_source3_page = response_sources[2].metadata["page"] + 1
|
85 |
+
new_history = history + [(message, response_answer)]
|
86 |
+
return qa_chain, gr.update(value=""), new_history, response_source1, response_source1_page, response_source2, response_source2_page, response_source3, response_source3_page
|
87 |
+
|
88 |
+
def demo():
|
89 |
+
with gr.Blocks(theme="base") as demo:
|
90 |
+
vector_db = gr.State()
|
91 |
+
qa_chain = gr.State()
|
92 |
+
collection_name = gr.State()
|
93 |
+
|
94 |
+
gr.Markdown(
|
95 |
+
"""<center><h2>PDF-based chatbot (powered by LangChain and open-source LLMs)</center></h2>
|
96 |
+
<h3>Ask any questions about your PDF documents, along with follow-ups</h3>
|
97 |
+
<b>Note:</b> This AI assistant performs retrieval-augmented generation from your PDF documents.
|
98 |
+
When generating answers, it takes past questions into account (via conversational memory), and includes document references for clarity purposes.
|
99 |
+
<br><b>Warning:</b> This space uses the free CPU Basic hardware from Hugging Face. Some steps and LLM models used below (free inference endpoints) can take some time to generate an output.<br>
|
100 |
+
""")
|
101 |
+
|
102 |
+
with gr.Tab("Step 1 - Document pre-processing"):
|
103 |
+
document = gr.Files(height=100, file_count="multiple", file_types=["pdf"], interactive=True, label="Upload your PDF documents (single or multiple)")
|
104 |
+
with gr.Row():
|
105 |
+
db_progress = gr.Textbox(label="Vector database initialization", value="None")
|
106 |
+
with gr.Row():
|
107 |
+
db_btn = gr.Button("Generate vector database...")
|
108 |
+
|
109 |
+
with gr.Tab("Step 2 - QA chain initialization"):
|
110 |
+
llm_btn = gr.Radio(["mistralai/Mistral-7B-Instruct-v0.2"], label="LLM models", value="mistralai/Mistral-7B-Instruct-v0.2", type="index", info="Choose your LLM model")
|
111 |
+
with gr.Row():
|
112 |
+
llm_progress = gr.Textbox(value="None", label="QA chain initialization")
|
113 |
+
with gr.Row():
|
114 |
+
qachain_btn = gr.Button("Initialize question-answering chain...")
|
115 |
+
|
116 |
+
with gr.Tab("Step 3 - Conversation with chatbot"):
|
117 |
+
chatbot = gr.Chatbot(height=300)
|
118 |
+
with gr.Row():
|
119 |
+
msg = gr.Textbox(placeholder="Type message", container=True)
|
120 |
+
with gr.Row():
|
121 |
+
submit_btn = gr.Button("Submit")
|
122 |
+
clear_btn = gr.ClearButton([msg, chatbot])
|
123 |
+
|
124 |
+
db_btn.click(initialize_database, inputs=[document], outputs=[vector_db, collection_name, db_progress])
|
125 |
+
qachain_btn.click(initialize_LLM, inputs=[llm_btn, vector_db], outputs=[qa_chain, llm_progress]).then(lambda:[None,"",0,"",0,"",0], inputs=None, outputs=[chatbot], queue=False)
|
126 |
+
|
127 |
+
msg.submit(conversation, inputs=[qa_chain, msg, chatbot], outputs=[qa_chain, msg, chatbot], queue=False)
|
128 |
+
submit_btn.click(conversation, inputs=[qa_chain, msg, chatbot], outputs=[qa_chain, msg, chatbot], queue=False)
|
129 |
+
clear_btn.click(lambda:[None,"",0,"",0,"",0], inputs=None, outputs=[chatbot], queue=False)
|
130 |
+
|
131 |
+
demo.queue().launch(debug=True)
|
132 |
+
|
133 |
+
if __name__ == "__main__":
|
134 |
+
demo()
|