afatuddin commited on
Commit
400dd32
·
verified ·
1 Parent(s): b4fa060

Delete app.py

Browse files
Files changed (1) hide show
  1. app.py +0 -196
app.py DELETED
@@ -1,196 +0,0 @@
1
- import streamlit as st
2
- from PyPDF2 import PdfReader
3
-
4
- # Import Document schema to hold text + metadata (page number)
5
- from langchain.schema import Document
6
- from langchain_text_splitters import RecursiveCharacterTextSplitter
7
- from langchain_community.vectorstores import FAISS
8
- from langchain_community.chat_models import ChatOllama
9
- from langchain_community.embeddings import OllamaEmbeddings
10
-
11
- from langchain.chains.question_answering import load_qa_chain
12
- from langchain_core.prompts import PromptTemplate
13
-
14
-
15
- @st.cache_data
16
- def get_pdf_text(pdf_docs):
17
- """
18
- Extracts text from PDFs and converts them into Document objects
19
- containing the page number and filename in metadata.
20
- """
21
- documents = []
22
- for pdf in pdf_docs:
23
- pdf_reader = PdfReader(pdf)
24
- pdf_filename = pdf.name
25
-
26
- for i, page in enumerate(pdf_reader.pages):
27
- text = page.extract_text()
28
- if text:
29
- # Create a Document object for every page
30
- # 'i + 1' ensures pages start at 1, not 0
31
- doc = Document(
32
- page_content=text,
33
- metadata={"page": i + 1, "source": pdf_filename}
34
- )
35
- documents.append(doc)
36
- return documents
37
-
38
-
39
- @st.cache_data
40
- def get_text_chunks(documents):
41
- """
42
- Splits Document objects into manageable chunks while preserving metadata.
43
- """
44
- text_splitter = RecursiveCharacterTextSplitter(
45
- chunk_size=2000, # Reduced size for better page granularity
46
- chunk_overlap=200
47
- )
48
- # split_documents (instead of split_text) keeps the metadata attached to chunks
49
- chunks = text_splitter.split_documents(documents)
50
- return chunks
51
-
52
-
53
- def get_vector_store(text_chunks):
54
- """
55
- Creates and saves a local FAISS vector store from Document chunks.
56
- """
57
- if not text_chunks:
58
- st.error("No text chunks found. Cannot create vector store.")
59
- return
60
-
61
- # Use the local Ollama embedding model
62
- embeddings = OllamaEmbeddings(model="nomic-embed-text")
63
-
64
- try:
65
- # from_documents handles the Document objects including metadata
66
- vector_store = FAISS.from_documents(documents=text_chunks, embedding=embeddings)
67
- vector_store.save_local("faiss_index")
68
- st.success("Vector store created and saved locally.")
69
- except Exception as e:
70
- st.error(f"Error creating vector store: {e}")
71
- print(f"Error creating vector store: {e}")
72
-
73
-
74
- def get_conversational_chain():
75
- """
76
- Creates a question-answering chain with a local Ollama model.
77
- """
78
- prompt_template = """
79
- You are a helpful and informative AI assistant. Your primary goal is to answer questions based on the provided context, which consists of user-uploaded PDF documents.
80
- Instructions:
81
- 1. First, thoroughly analyze the content of the user-uploaded PDFs to extract relevant information for answering the question.
82
- 2. If the answer can be fully derived from the PDF content, provide it directly and ensure accuracy and completeness.
83
- 3. If you are uncertain about the answer or if the PDFs and your knowledge do not provide enough information to form a conclusive answer, state clearly that the information is not sufficient to provide a definitive answer.
84
- 4. Avoid fabricating information or making assumptions that cannot be supported by the PDF content or reliable external knowledge.
85
-
86
- **Important:** If the user prompt is not related to the PDF content,Just say "I don't know this answer".
87
- **Important:** If the user prompt is not related to the PDF content,Never say the answer if you know then still dont say.
88
-
89
- Context:\n{context}
90
- Question: \n{question}
91
- Answer:
92
- """
93
-
94
-
95
- model = ChatOllama(model="gemma3:1b", temperature=0.3)
96
-
97
- prompt = PromptTemplate(
98
- template=prompt_template,
99
- input_variables=["context", "question"]
100
- )
101
- chain = load_qa_chain(model, chain_type="stuff", prompt=prompt)
102
- return chain
103
-
104
-
105
- def load_faiss_index():
106
- """Loads the local FAISS index."""
107
- try:
108
- embeddings = OllamaEmbeddings(model="nomic-embed-text")
109
- db = FAISS.load_local(
110
- "faiss_index",
111
- embeddings,
112
- allow_dangerous_deserialization=True
113
- )
114
- return db
115
- except Exception as e:
116
- st.error(f"Failed to load FAISS index: {e}")
117
- return None
118
-
119
- def user_input(user_question, db):
120
- """
121
- Handles user question, performs similarity search, gets response,
122
- and extracts source page numbers programmatically.
123
- """
124
- if db is None:
125
- st.warning("Vector store is not loaded. Please process PDFs first.")
126
- return
127
-
128
- try:
129
- # 1. Retrieve relevant chunks (Docs)
130
- docs = db.similarity_search(user_question)
131
-
132
- # 2. Generate Answer using LLM
133
- chain = get_conversational_chain()
134
- response = chain(
135
- {"input_documents": docs, "question": user_question},
136
- return_only_outputs=True
137
- )
138
-
139
- # 3. Display Answer
140
- st.write("Reply: ", response["output_text"])
141
-
142
- # 4. Display Sources (Page Numbers)
143
- # We use a set to avoid duplicate page numbers if multiple chunks come from the same page
144
- sources = set()
145
- for doc in docs:
146
- page_num = doc.metadata.get("page", "Unknown")
147
- file_name = doc.metadata.get("source", "Unknown")
148
- sources.add(f"{file_name} (Page {page_num})")
149
-
150
- # Display the sources in an expandable box
151
- with st.expander("📚 View Source Pages"):
152
- for source in sources:
153
- st.markdown(f"- {source}")
154
-
155
- except Exception as e:
156
- st.error(f"Error processing your question: {e}")
157
- print(f"Error processing your question: {e}")
158
-
159
-
160
- def main():
161
- st.set_page_config(page_title="Chat PDF (Local)")
162
- st.header("Chat with PDF using Gemma3:1b 🤖")
163
-
164
- if "faiss_db" not in st.session_state:
165
- st.session_state.faiss_db = None
166
-
167
- with st.sidebar:
168
- st.title("Menu:")
169
- pdf_docs = st.file_uploader(
170
- "Upload your PDF Files and Click on the Process Button",
171
- accept_multiple_files=True
172
- )
173
-
174
- if st.button("Process PDFs"):
175
- if pdf_docs:
176
- with st.spinner("Processing PDF files..."):
177
- # Pass document objects now, not raw text
178
- raw_documents = get_pdf_text(pdf_docs)
179
- text_chunks = get_text_chunks(raw_documents)
180
- get_vector_store(text_chunks)
181
-
182
- st.session_state.faiss_db = load_faiss_index()
183
- else:
184
- st.warning("Please upload at least one PDF file.")
185
-
186
- user_question = st.text_input("Ask a Question from the PDF Files")
187
-
188
- if user_question:
189
- if st.session_state.faiss_db is not None:
190
- user_input(user_question, st.session_state.faiss_db)
191
- else:
192
- st.warning("Please upload and process at least one PDF first.")
193
-
194
-
195
- if __name__ == "__main__":
196
- main()