Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,11 +1,10 @@
|
|
1 |
import os
|
2 |
import streamlit as st
|
3 |
import pdfplumber
|
4 |
-
from concurrent.futures import ThreadPoolExecutor
|
5 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
6 |
from langchain.embeddings import HuggingFaceEmbeddings
|
7 |
from langchain.vectorstores import FAISS
|
8 |
-
from transformers import pipeline
|
9 |
|
10 |
# Set up the page configuration
|
11 |
st.set_page_config(page_title="RAG-based PDF Chat", layout="centered", page_icon="π")
|
@@ -13,29 +12,42 @@ st.set_page_config(page_title="RAG-based PDF Chat", layout="centered", page_icon
|
|
13 |
# Load the summarization pipeline model
|
14 |
@st.cache_resource
|
15 |
def load_summarization_pipeline():
|
16 |
-
|
17 |
-
return summarizer
|
18 |
|
19 |
summarizer = load_summarization_pipeline()
|
20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
# Split text into manageable chunks
|
22 |
@st.cache_data
|
23 |
def get_text_chunks(text):
|
24 |
text_splitter = RecursiveCharacterTextSplitter(chunk_size=10000, chunk_overlap=1000)
|
25 |
-
|
26 |
-
return chunks
|
27 |
|
28 |
# Initialize embedding function
|
29 |
embedding_function = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
|
30 |
|
31 |
-
# Create a FAISS vector store with embeddings
|
32 |
@st.cache_resource
|
33 |
def load_or_create_vector_store(text_chunks):
|
34 |
-
if
|
35 |
-
st.error("No valid text chunks found to create a vector store. Please check your PDF files.")
|
36 |
-
return None
|
37 |
-
vector_store = FAISS.from_texts(text_chunks, embedding=embedding_function)
|
38 |
-
return vector_store
|
39 |
|
40 |
# Helper function to process a single PDF
|
41 |
def process_single_pdf(file_path):
|
@@ -50,65 +62,47 @@ def process_single_pdf(file_path):
|
|
50 |
st.error(f"Failed to read PDF: {file_path} - {e}")
|
51 |
return text
|
52 |
|
53 |
-
#
|
54 |
def load_pdfs_with_progress(folder_path):
|
|
|
|
|
|
|
|
|
55 |
all_text = ""
|
56 |
pdf_files = [os.path.join(folder_path, filename) for filename in os.listdir(folder_path) if filename.endswith('.pdf')]
|
57 |
-
|
58 |
-
|
59 |
-
if num_files == 0:
|
60 |
st.error("No PDF files found in the specified folder.")
|
61 |
-
|
62 |
-
st.session_state['loading'] = False
|
63 |
-
return
|
64 |
|
65 |
-
# Title for the progress bar
|
66 |
st.markdown("### Loading data...")
|
67 |
progress_bar = st.progress(0)
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
progress_bar.empty() # Remove the progress bar when done
|
81 |
-
status_text.text("Document loading completed!") # Show completion message
|
82 |
-
|
83 |
-
if all_text:
|
84 |
-
text_chunks = get_text_chunks(all_text)
|
85 |
-
vector_store = load_or_create_vector_store(text_chunks)
|
86 |
-
st.session_state['vector_store'] = vector_store
|
87 |
-
else:
|
88 |
-
st.session_state['vector_store'] = None
|
89 |
-
|
90 |
-
st.session_state['loading'] = False # Mark loading as complete
|
91 |
-
|
92 |
-
# Generate summary based on the retrieved text
|
93 |
-
def generate_summary_with_huggingface(query, retrieved_text):
|
94 |
-
summarization_input = f"{query} Related information:{retrieved_text}"
|
95 |
-
max_input_length = 1024
|
96 |
-
summarization_input = summarization_input[:max_input_length]
|
97 |
summary = summarizer(summarization_input, max_length=500, min_length=50, do_sample=False)
|
98 |
return summary[0]["summary_text"]
|
99 |
|
100 |
-
#
|
101 |
-
def
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
|
|
|
|
108 |
|
109 |
# Main function to run the Streamlit app
|
110 |
def main():
|
111 |
-
# Use HTML to style the title with a larger font size
|
112 |
st.markdown(
|
113 |
"""
|
114 |
<h1 style="font-size:30px; text-align: center;">
|
@@ -118,23 +112,24 @@ def main():
|
|
118 |
unsafe_allow_html=True
|
119 |
)
|
120 |
|
121 |
-
|
122 |
-
|
123 |
-
st.session_state[
|
124 |
-
|
125 |
|
|
|
126 |
user_question = st.text_input("Ask a Question:", placeholder="Type your question here...")
|
127 |
|
128 |
-
|
129 |
-
|
130 |
|
131 |
-
if st.button("Get Response"):
|
132 |
-
|
133 |
-
st.
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
|
139 |
if __name__ == "__main__":
|
140 |
main()
|
|
|
1 |
import os
|
2 |
import streamlit as st
|
3 |
import pdfplumber
|
|
|
4 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
5 |
from langchain.embeddings import HuggingFaceEmbeddings
|
6 |
from langchain.vectorstores import FAISS
|
7 |
+
from transformers import pipeline, M2M100ForConditionalGeneration, AutoTokenizer
|
8 |
|
9 |
# Set up the page configuration
|
10 |
st.set_page_config(page_title="RAG-based PDF Chat", layout="centered", page_icon="π")
|
|
|
12 |
# Load the summarization pipeline model
|
13 |
@st.cache_resource
|
14 |
def load_summarization_pipeline():
|
15 |
+
return pipeline("summarization", model="facebook/bart-large-cnn")
|
|
|
16 |
|
17 |
summarizer = load_summarization_pipeline()
|
18 |
|
19 |
+
# Load the translation model
|
20 |
+
@st.cache_resource
|
21 |
+
def load_translation_model():
|
22 |
+
model = M2M100ForConditionalGeneration.from_pretrained("alirezamsh/small100")
|
23 |
+
tokenizer = AutoTokenizer.from_pretrained("alirezamsh/small100")
|
24 |
+
return model, tokenizer
|
25 |
+
|
26 |
+
translation_model, translation_tokenizer = load_translation_model()
|
27 |
+
|
28 |
+
# Define available languages for translation
|
29 |
+
LANGUAGES = {
|
30 |
+
"English": "en",
|
31 |
+
"French": "fr",
|
32 |
+
"Spanish": "es",
|
33 |
+
"Chinese": "zh",
|
34 |
+
"Hindi": "hi",
|
35 |
+
"Urdu": "ur",
|
36 |
+
}
|
37 |
+
|
38 |
# Split text into manageable chunks
|
39 |
@st.cache_data
|
40 |
def get_text_chunks(text):
|
41 |
text_splitter = RecursiveCharacterTextSplitter(chunk_size=10000, chunk_overlap=1000)
|
42 |
+
return text_splitter.split_text(text)
|
|
|
43 |
|
44 |
# Initialize embedding function
|
45 |
embedding_function = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
|
46 |
|
47 |
+
# Create a FAISS vector store with embeddings
|
48 |
@st.cache_resource
|
49 |
def load_or_create_vector_store(text_chunks):
|
50 |
+
return FAISS.from_texts(text_chunks, embedding=embedding_function) if text_chunks else None
|
|
|
|
|
|
|
|
|
51 |
|
52 |
# Helper function to process a single PDF
|
53 |
def process_single_pdf(file_path):
|
|
|
62 |
st.error(f"Failed to read PDF: {file_path} - {e}")
|
63 |
return text
|
64 |
|
65 |
+
# Load PDFs with progress display
|
66 |
def load_pdfs_with_progress(folder_path):
|
67 |
+
if not os.path.exists(folder_path):
|
68 |
+
st.error(f"The folder '{folder_path}' does not exist. Please create it and add PDF files.")
|
69 |
+
return None
|
70 |
+
|
71 |
all_text = ""
|
72 |
pdf_files = [os.path.join(folder_path, filename) for filename in os.listdir(folder_path) if filename.endswith('.pdf')]
|
73 |
+
if not pdf_files:
|
|
|
|
|
74 |
st.error("No PDF files found in the specified folder.")
|
75 |
+
return None
|
|
|
|
|
76 |
|
|
|
77 |
st.markdown("### Loading data...")
|
78 |
progress_bar = st.progress(0)
|
79 |
+
|
80 |
+
for i, file_path in enumerate(pdf_files):
|
81 |
+
all_text += process_single_pdf(file_path)
|
82 |
+
progress_bar.progress((i + 1) / len(pdf_files))
|
83 |
+
|
84 |
+
progress_bar.empty()
|
85 |
+
return load_or_create_vector_store(get_text_chunks(all_text)) if all_text else None
|
86 |
+
|
87 |
+
# Generate summary based on retrieved text
|
88 |
+
def generate_summary(query, retrieved_text):
|
89 |
+
summarization_input = f"{query} Related information:{retrieved_text}"[:1024]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
90 |
summary = summarizer(summarization_input, max_length=500, min_length=50, do_sample=False)
|
91 |
return summary[0]["summary_text"]
|
92 |
|
93 |
+
# Translate text to selected language
|
94 |
+
def translate_text(text, target_lang_code):
|
95 |
+
# Set the target language
|
96 |
+
translation_tokenizer.src_lang = "en" # assuming the original language is English
|
97 |
+
translation_tokenizer.tgt_lang = target_lang_code # dynamically use the selected target language code
|
98 |
+
|
99 |
+
# Tokenize the text and generate the translation
|
100 |
+
encoded_text = translation_tokenizer(text, return_tensors="pt")
|
101 |
+
generated_tokens = translation_model.generate(**encoded_text, forced_bos_token_id=translation_tokenizer.lang_code_to_id[target_lang_code])
|
102 |
+
return translation_tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
|
103 |
|
104 |
# Main function to run the Streamlit app
|
105 |
def main():
|
|
|
106 |
st.markdown(
|
107 |
"""
|
108 |
<h1 style="font-size:30px; text-align: center;">
|
|
|
112 |
unsafe_allow_html=True
|
113 |
)
|
114 |
|
115 |
+
if "vector_store" not in st.session_state:
|
116 |
+
st.session_state["vector_store"] = load_pdfs_with_progress('documents1')
|
117 |
+
if st.session_state["vector_store"] is None:
|
118 |
+
return
|
119 |
|
120 |
+
# Prompt input
|
121 |
user_question = st.text_input("Ask a Question:", placeholder="Type your question here...")
|
122 |
|
123 |
+
# Language selection dropdown
|
124 |
+
selected_language = st.selectbox("Select output language:", list(LANGUAGES.keys()))
|
125 |
|
126 |
+
if user_question and st.button("Get Response"):
|
127 |
+
with st.spinner("Generating response..."):
|
128 |
+
docs = st.session_state["vector_store"].similarity_search(user_question)
|
129 |
+
context_text = " ".join([doc.page_content for doc in docs])
|
130 |
+
answer = generate_summary(user_question, context_text)
|
131 |
+
translated_answer = translate_text(answer, LANGUAGES[selected_language])
|
132 |
+
st.markdown(f"**π€ AI ({selected_language}):** {translated_answer}")
|
133 |
|
134 |
if __name__ == "__main__":
|
135 |
main()
|