Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,164 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import os
|
3 |
+
from langchain.document_loaders import PyPDFLoader
|
4 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
5 |
+
from langchain.vectorstores import Chroma
|
6 |
+
from langchain.chains import ConversationalRetrievalChain
|
7 |
+
from langchain.embeddings import HuggingFaceEmbeddings
|
8 |
+
from langchain.llms import HuggingFaceHub
|
9 |
+
from langchain.memory import ConversationBufferMemory
|
10 |
+
import chromadb
|
11 |
+
from transformers import AutoTokenizer
|
12 |
+
import transformers
|
13 |
+
import torch
|
14 |
+
|
15 |
+
# Constants and configuration
|
16 |
+
list_llm = [
|
17 |
+
"mistralai/Mixtral-8x7B-Instruct-v0.1", "mistralai/Mistral-7B-Instruct-v0.2",
|
18 |
+
"mistralai/Mistral-7B-Instruct-v0.1", "HuggingFaceH4/zephyr-7b-beta",
|
19 |
+
"meta-llama/Llama-2-7b-chat-hf", "microsoft/phi-2",
|
20 |
+
"TinyLlama/TinyLlama-1.1B-Chat-v1.0", "mosaicml/mpt-7b-instruct",
|
21 |
+
"tiiuae/falcon-7b-instruct", "google/flan-t5-xxl"
|
22 |
+
]
|
23 |
+
list_llm_simple = [os.path.basename(llm) for llm in list_llm]
|
24 |
+
|
25 |
+
# Function placeholders (actual function implementations from the original script)
|
26 |
+
def load_doc(list_file_path, chunk_size, chunk_overlap):
|
27 |
+
loaders = [PyPDFLoader(x) for x in list_file_path]
|
28 |
+
pages = []
|
29 |
+
for loader in loaders:
|
30 |
+
pages.extend(loader.load())
|
31 |
+
text_splitter = RecursiveCharacterTextSplitter(
|
32 |
+
chunk_size = chunk_size,
|
33 |
+
chunk_overlap = chunk_overlap)
|
34 |
+
doc_splits = text_splitter.split_documents(pages)
|
35 |
+
return doc_splits
|
36 |
+
|
37 |
+
def create_db(splits, collection_name):
|
38 |
+
embedding = HuggingFaceEmbeddings()
|
39 |
+
new_client = chromadb.EphemeralClient()
|
40 |
+
vectordb = Chroma.from_documents(
|
41 |
+
documents=splits,
|
42 |
+
embedding=embedding,
|
43 |
+
client=new_client,
|
44 |
+
collection_name=collection_name,
|
45 |
+
)
|
46 |
+
return vectordb
|
47 |
+
|
48 |
+
def load_db():
|
49 |
+
embedding = HuggingFaceEmbeddings()
|
50 |
+
vectordb = Chroma(embedding_function=embedding)
|
51 |
+
return vectordb
|
52 |
+
|
53 |
+
def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
|
54 |
+
if llm_model == "mistralai/Mixtral-8x7B-Instruct-v0.1":
|
55 |
+
llm = HuggingFaceHub(
|
56 |
+
repo_id=llm_model,
|
57 |
+
model_kwargs={"temperature": temperature, "max_new_tokens": max_tokens, "top_k": top_k, "load_in_8bit": True}
|
58 |
+
)
|
59 |
+
else:
|
60 |
+
llm = HuggingFaceHub(
|
61 |
+
repo_id=llm_model,
|
62 |
+
model_kwargs={"temperature": temperature, "max_new_tokens": max_tokens, "top_k": top_k}
|
63 |
+
)
|
64 |
+
memory = ConversationBufferMemory(
|
65 |
+
memory_key="chat_history",
|
66 |
+
output_key='answer',
|
67 |
+
return_messages=True
|
68 |
+
)
|
69 |
+
retriever = vector_db.as_retriever()
|
70 |
+
qa_chain = ConversationalRetrievalChain.from_llm(
|
71 |
+
llm,
|
72 |
+
retriever=retriever,
|
73 |
+
chain_type="stuff",
|
74 |
+
memory=memory,
|
75 |
+
return_source_documents=True,
|
76 |
+
return_generated_question=False,
|
77 |
+
)
|
78 |
+
return qa_chain
|
79 |
+
|
80 |
+
def initialize_database(list_file_obj, chunk_size, chunk_overlap):
|
81 |
+
list_file_path = [x.name for x in list_file_obj if x is not None]
|
82 |
+
collection_name = os.path.basename(list_file_path[0]).replace(" ","-")[:50]
|
83 |
+
doc_splits = load_doc(list_file_path, chunk_size, chunk_overlap)
|
84 |
+
vector_db = create_db(doc_splits, collection_name)
|
85 |
+
return vector_db, collection_name, "Complete!"
|
86 |
+
|
87 |
+
def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db):
|
88 |
+
llm_name = list_llm[llm_option]
|
89 |
+
qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, top_k, vector_db)
|
90 |
+
return qa_chain, "Complete!"
|
91 |
+
|
92 |
+
def format_chat_history(message, chat_history):
|
93 |
+
formatted_chat_history = []
|
94 |
+
for user_message, bot_message in chat_history:
|
95 |
+
formatted_chat_history.append(f"User: {user_message}")
|
96 |
+
formatted_chat_history.append(f"Assistant: {bot_message}")
|
97 |
+
return formatted_chat_history
|
98 |
+
|
99 |
+
def conversation(qa_chain, message, history):
|
100 |
+
formatted_chat_history = format_chat_history(message, history)
|
101 |
+
response = qa_chain({"question": message, "chat_history": formatted_chat_history})
|
102 |
+
response_answer = response["answer"]
|
103 |
+
response_sources = response["source_documents"]
|
104 |
+
response_source1 = response_sources[0].page_content.strip()
|
105 |
+
response_source2 = response_sources[1].page_content.strip()
|
106 |
+
response_source1_page = response_sources[0].metadata["page"] + 1
|
107 |
+
response_source2_page = response_sources[1].metadata["page"] + 1
|
108 |
+
|
109 |
+
new_history = history + [(message, response_answer)]
|
110 |
+
return qa_chain, gr.update(value=""), new_history, response_source1, response_source1_page, response_source2, response_source2_page
|
111 |
+
|
112 |
+
def upload_file(file_obj):
|
113 |
+
list_file_path = [file.name for file in file_obj]
|
114 |
+
return list_file_path
|
115 |
+
|
116 |
+
|
117 |
+
def gradio_ui():
|
118 |
+
with gr.Blocks(theme="base") as demo:
|
119 |
+
# States
|
120 |
+
vector_db, qa_chain, collection_name = gr.State(), gr.State(), gr.State()
|
121 |
+
db_progress, llm_progress = gr.Textbox(), gr.Textbox()
|
122 |
+
chatbot, doc_source1, source1_page, doc_source2, source2_page = gr.Chatbot(), gr.Textbox(), gr.Number(), gr.Textbox(), gr.Number()
|
123 |
+
msg = gr.Textbox(placeholder="Type message")
|
124 |
+
|
125 |
+
with gr.Tabs():
|
126 |
+
# Tab 1: Document Pre-processing
|
127 |
+
with gr.Tab("Step 1 - Document Pre-processing"):
|
128 |
+
with gr.Row():
|
129 |
+
document = gr.File(label="Upload your PDF document", file_types=["pdf"])
|
130 |
+
with gr.Row():
|
131 |
+
chunk_size = gr.Slider(minimum=100, maximum=1000, value=600, step=50, label="Chunk size", interactive=True)
|
132 |
+
chunk_overlap = gr.Slider(minimum=10, maximum=200, value=50, step=10, label="Chunk overlap", interactive=True)
|
133 |
+
with gr.Row():
|
134 |
+
db_init_btn = gr.Button("Initialize Vector Database")
|
135 |
+
|
136 |
+
# Tab 2: QA Chain Initialization
|
137 |
+
with gr.Tab("Step 2 - QA Chain Initialization"):
|
138 |
+
with gr.Row():
|
139 |
+
llm_selection = gr.Radio(list_llm_simple, label="Choose LLM Model", value=list_llm_simple[0])
|
140 |
+
with gr.Row():
|
141 |
+
temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.7, step=0.1, label="Temperature", interactive=True)
|
142 |
+
max_tokens = gr.Slider(minimum=64, maximum=1024, value=256, step=64, label="Max Tokens", interactive=True)
|
143 |
+
top_k = gr.Slider(minimum=1, maximum=10, value=3, step=1, label="Top K", interactive=True)
|
144 |
+
with gr.Row():
|
145 |
+
qa_init_btn = gr.Button("Initialize QA Chain")
|
146 |
+
|
147 |
+
# Tab 3: Conversation with Chatbot
|
148 |
+
with gr.Tab("Step 3 - Conversation with Chatbot"):
|
149 |
+
chat_history = gr.State()
|
150 |
+
with gr.Row():
|
151 |
+
chatbot
|
152 |
+
with gr.Row():
|
153 |
+
msg
|
154 |
+
submit_btn = gr.Button("Submit")
|
155 |
+
|
156 |
+
# Handlers
|
157 |
+
db_init_btn.click(initialize_database, inputs=[document, chunk_size, chunk_overlap], outputs=[vector_db, collection_name, db_progress])
|
158 |
+
qa_init_btn.click(initialize_LLM, inputs=[llm_selection, temperature, max_tokens, top_k, vector_db], outputs=[qa_chain, llm_progress])
|
159 |
+
submit_btn.click(conversation, inputs=[qa_chain, msg, chat_history], outputs=[qa_chain, msg, chatbot, doc_source1, source1_page, doc_source2, source2_page])
|
160 |
+
|
161 |
+
return demo
|
162 |
+
|
163 |
+
if __name__ == "__main__":
|
164 |
+
gradio_ui().launch()
|