Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,278 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
Streamlit application for PDF-based Retrieval-Augmented Generation (RAG) using Ollama + LangChain.
|
3 |
+
|
4 |
+
This application allows users to upload a PDF, process it,
|
5 |
+
and then ask questions about the content using a selected language model.
|
6 |
+
"""
|
7 |
+
|
8 |
+
import streamlit as st
|
9 |
+
import logging
|
10 |
+
import os
|
11 |
+
import tempfile
|
12 |
+
import shutil
|
13 |
+
import pdfplumber
|
14 |
+
import ollama
|
15 |
+
|
16 |
+
from langchain_community.document_loaders import UnstructuredPDFLoader
|
17 |
+
from langchain_community.embeddings import OllamaEmbeddings
|
18 |
+
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
19 |
+
from langchain_community.vectorstores import Chroma
|
20 |
+
from langchain.prompts import ChatPromptTemplate, PromptTemplate
|
21 |
+
from langchain_core.output_parsers import StrOutputParser
|
22 |
+
from langchain_community.chat_models import ChatOllama
|
23 |
+
from langchain_core.runnables import RunnablePassthrough
|
24 |
+
from langchain.retrievers.multi_query import MultiQueryRetriever
|
25 |
+
from typing import List, Tuple, Dict, Any, Optional
|
26 |
+
|
27 |
+
# Streamlit page configuration
|
28 |
+
st.set_page_config(
|
29 |
+
page_title="Ollama PDF RAG Streamlit UI",
|
30 |
+
page_icon="π",
|
31 |
+
layout="wide",
|
32 |
+
initial_sidebar_state="collapsed",
|
33 |
+
)
|
34 |
+
|
35 |
+
# Logging configuration
|
36 |
+
logging.basicConfig(
|
37 |
+
level=logging.INFO,
|
38 |
+
format="%(asctime)s - %(levelname)s - %(message)s",
|
39 |
+
datefmt="%Y-%m-%d %H:%M:%S",
|
40 |
+
)
|
41 |
+
|
42 |
+
logger = logging.getLogger(__name__)
|
43 |
+
|
44 |
+
|
45 |
+
@st.cache_resource(show_spinner=True)
|
46 |
+
def extract_model_names(
|
47 |
+
models_info: Dict[str, List[Dict[str, Any]]],
|
48 |
+
) -> Tuple[str, ...]:
|
49 |
+
"""
|
50 |
+
Extract model names from the provided models information.
|
51 |
+
|
52 |
+
Args:
|
53 |
+
models_info (Dict[str, List[Dict[str, Any]]]): Dictionary containing information about available models.
|
54 |
+
|
55 |
+
Returns:
|
56 |
+
Tuple[str, ...]: A tuple of model names.
|
57 |
+
"""
|
58 |
+
logger.info("Extracting model names from models_info")
|
59 |
+
model_names = tuple(model["name"] for model in models_info["models"])
|
60 |
+
logger.info(f"Extracted model names: {model_names}")
|
61 |
+
return model_names
|
62 |
+
|
63 |
+
|
64 |
+
def create_vector_db(file_upload) -> Chroma:
|
65 |
+
"""
|
66 |
+
Create a vector database from an uploaded PDF file.
|
67 |
+
|
68 |
+
Args:
|
69 |
+
file_upload (st.UploadedFile): Streamlit file upload object containing the PDF.
|
70 |
+
|
71 |
+
Returns:
|
72 |
+
Chroma: A vector store containing the processed document chunks.
|
73 |
+
"""
|
74 |
+
logger.info(f"Creating vector DB from file upload: {file_upload.name}")
|
75 |
+
temp_dir = tempfile.mkdtemp()
|
76 |
+
|
77 |
+
path = os.path.join(temp_dir, file_upload.name)
|
78 |
+
with open(path, "wb") as f:
|
79 |
+
f.write(file_upload.getvalue())
|
80 |
+
logger.info(f"File saved to temporary path: {path}")
|
81 |
+
loader = UnstructuredPDFLoader(path)
|
82 |
+
data = loader.load()
|
83 |
+
|
84 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=7500, chunk_overlap=100)
|
85 |
+
chunks = text_splitter.split_documents(data)
|
86 |
+
logger.info("Document split into chunks")
|
87 |
+
|
88 |
+
embeddings = OllamaEmbeddings(model="nomic-embed-text", show_progress=True)
|
89 |
+
vector_db = Chroma.from_documents(
|
90 |
+
documents=chunks, embedding=embeddings, collection_name="myRAG"
|
91 |
+
)
|
92 |
+
logger.info("Vector DB created")
|
93 |
+
|
94 |
+
shutil.rmtree(temp_dir)
|
95 |
+
logger.info(f"Temporary directory {temp_dir} removed")
|
96 |
+
return vector_db
|
97 |
+
|
98 |
+
|
99 |
+
def process_question(question: str, vector_db: Chroma, selected_model: str) -> str:
|
100 |
+
"""
|
101 |
+
Process a user question using the vector database and selected language model.
|
102 |
+
|
103 |
+
Args:
|
104 |
+
question (str): The user's question.
|
105 |
+
vector_db (Chroma): The vector database containing document embeddings.
|
106 |
+
selected_model (str): The name of the selected language model.
|
107 |
+
|
108 |
+
Returns:
|
109 |
+
str: The generated response to the user's question.
|
110 |
+
"""
|
111 |
+
logger.info(f"""Processing question: {
|
112 |
+
question} using model: {selected_model}""")
|
113 |
+
llm = ChatOllama(model=selected_model, temperature=0)
|
114 |
+
QUERY_PROMPT = PromptTemplate(
|
115 |
+
input_variables=["question"],
|
116 |
+
template="""You are an AI language model assistant. Your task is to generate 3
|
117 |
+
different versions of the given user question to retrieve relevant documents from
|
118 |
+
a vector database. By generating multiple perspectives on the user question, your
|
119 |
+
goal is to help the user overcome some of the limitations of the distance-based
|
120 |
+
similarity search. Provide these alternative questions separated by newlines.
|
121 |
+
Original question: {question}""",
|
122 |
+
)
|
123 |
+
|
124 |
+
retriever = MultiQueryRetriever.from_llm(
|
125 |
+
vector_db.as_retriever(), llm, prompt=QUERY_PROMPT
|
126 |
+
)
|
127 |
+
|
128 |
+
template = """Answer the question based ONLY on the following context:
|
129 |
+
{context}
|
130 |
+
Question: {question}
|
131 |
+
If you don't know the answer, just say that you don't know, don't try to make up an answer.
|
132 |
+
Only provide the answer from the {context}, nothing else.
|
133 |
+
Add snippets of the context you used to answer the question.
|
134 |
+
"""
|
135 |
+
|
136 |
+
prompt = ChatPromptTemplate.from_template(template)
|
137 |
+
|
138 |
+
chain = (
|
139 |
+
{"context": retriever, "question": RunnablePassthrough()}
|
140 |
+
| prompt
|
141 |
+
| llm
|
142 |
+
| StrOutputParser()
|
143 |
+
)
|
144 |
+
|
145 |
+
response = chain.invoke(question)
|
146 |
+
logger.info("Question processed and response generated")
|
147 |
+
return response
|
148 |
+
|
149 |
+
|
150 |
+
@st.cache_data
|
151 |
+
def extract_all_pages_as_images(file_upload) -> List[Any]:
|
152 |
+
"""
|
153 |
+
Extract all pages from a PDF file as images.
|
154 |
+
|
155 |
+
Args:
|
156 |
+
file_upload (st.UploadedFile): Streamlit file upload object containing the PDF.
|
157 |
+
|
158 |
+
Returns:
|
159 |
+
List[Any]: A list of image objects representing each page of the PDF.
|
160 |
+
"""
|
161 |
+
logger.info(f"""Extracting all pages as images from file: {
|
162 |
+
file_upload.name}""")
|
163 |
+
pdf_pages = []
|
164 |
+
with pdfplumber.open(file_upload) as pdf:
|
165 |
+
pdf_pages = [page.to_image().original for page in pdf.pages]
|
166 |
+
logger.info("PDF pages extracted as images")
|
167 |
+
return pdf_pages
|
168 |
+
|
169 |
+
|
170 |
+
def delete_vector_db(vector_db: Optional[Chroma]) -> None:
|
171 |
+
"""
|
172 |
+
Delete the vector database and clear related session state.
|
173 |
+
|
174 |
+
Args:
|
175 |
+
vector_db (Optional[Chroma]): The vector database to be deleted.
|
176 |
+
"""
|
177 |
+
logger.info("Deleting vector DB")
|
178 |
+
if vector_db is not None:
|
179 |
+
vector_db.delete_collection()
|
180 |
+
st.session_state.pop("pdf_pages", None)
|
181 |
+
st.session_state.pop("file_upload", None)
|
182 |
+
st.session_state.pop("vector_db", None)
|
183 |
+
st.success("Collection and temporary files deleted successfully.")
|
184 |
+
logger.info("Vector DB and related session state cleared")
|
185 |
+
st.rerun()
|
186 |
+
else:
|
187 |
+
st.error("No vector database found to delete.")
|
188 |
+
logger.warning("Attempted to delete vector DB, but none was found")
|
189 |
+
|
190 |
+
|
191 |
+
def main() -> None:
|
192 |
+
"""
|
193 |
+
Main function to run the Streamlit application.
|
194 |
+
|
195 |
+
This function sets up the user interface, handles file uploads,
|
196 |
+
processes user queries, and displays results.
|
197 |
+
"""
|
198 |
+
st.subheader("π§ Ollama PDF RAG playground", divider="gray", anchor=False)
|
199 |
+
|
200 |
+
models_info = ollama.list()
|
201 |
+
available_models = extract_model_names(models_info)
|
202 |
+
|
203 |
+
col1, col2 = st.columns([1.5, 2])
|
204 |
+
|
205 |
+
if "messages" not in st.session_state:
|
206 |
+
st.session_state["messages"] = []
|
207 |
+
|
208 |
+
if "vector_db" not in st.session_state:
|
209 |
+
st.session_state["vector_db"] = None
|
210 |
+
|
211 |
+
if available_models:
|
212 |
+
selected_model = col2.selectbox(
|
213 |
+
"Pick a model available locally on your system β", available_models
|
214 |
+
)
|
215 |
+
|
216 |
+
file_upload = col1.file_uploader(
|
217 |
+
"Upload a PDF file β", type="pdf", accept_multiple_files=False
|
218 |
+
)
|
219 |
+
|
220 |
+
if file_upload:
|
221 |
+
st.session_state["file_upload"] = file_upload
|
222 |
+
if st.session_state["vector_db"] is None:
|
223 |
+
st.session_state["vector_db"] = create_vector_db(file_upload)
|
224 |
+
pdf_pages = extract_all_pages_as_images(file_upload)
|
225 |
+
st.session_state["pdf_pages"] = pdf_pages
|
226 |
+
|
227 |
+
zoom_level = col1.slider(
|
228 |
+
"Zoom Level", min_value=100, max_value=1000, value=700, step=50
|
229 |
+
)
|
230 |
+
|
231 |
+
with col1:
|
232 |
+
with st.container(height=410, border=True):
|
233 |
+
for page_image in pdf_pages:
|
234 |
+
st.image(page_image, width=zoom_level)
|
235 |
+
|
236 |
+
delete_collection = col1.button("β οΈ Delete collection", type="secondary")
|
237 |
+
|
238 |
+
if delete_collection:
|
239 |
+
delete_vector_db(st.session_state["vector_db"])
|
240 |
+
|
241 |
+
with col2:
|
242 |
+
message_container = st.container(height=500, border=True)
|
243 |
+
|
244 |
+
for message in st.session_state["messages"]:
|
245 |
+
avatar = "π€" if message["role"] == "assistant" else "π"
|
246 |
+
with message_container.chat_message(message["role"], avatar=avatar):
|
247 |
+
st.markdown(message["content"])
|
248 |
+
|
249 |
+
if prompt := st.chat_input("Enter a prompt here..."):
|
250 |
+
try:
|
251 |
+
st.session_state["messages"].append({"role": "user", "content": prompt})
|
252 |
+
message_container.chat_message("user", avatar="π").markdown(prompt)
|
253 |
+
|
254 |
+
with message_container.chat_message("assistant", avatar="π€"):
|
255 |
+
with st.spinner(":green[processing...]"):
|
256 |
+
if st.session_state["vector_db"] is not None:
|
257 |
+
response = process_question(
|
258 |
+
prompt, st.session_state["vector_db"], selected_model
|
259 |
+
)
|
260 |
+
st.markdown(response)
|
261 |
+
else:
|
262 |
+
st.warning("Please upload a PDF file first.")
|
263 |
+
|
264 |
+
if st.session_state["vector_db"] is not None:
|
265 |
+
st.session_state["messages"].append(
|
266 |
+
{"role": "assistant", "content": response}
|
267 |
+
)
|
268 |
+
|
269 |
+
except Exception as e:
|
270 |
+
st.error(e, icon="βοΈ")
|
271 |
+
logger.error(f"Error processing prompt: {e}")
|
272 |
+
else:
|
273 |
+
if st.session_state["vector_db"] is None:
|
274 |
+
st.warning("Upload a PDF file to begin chat...")
|
275 |
+
|
276 |
+
|
277 |
+
if __name__ == "__main__":
|
278 |
+
main()
|