Optimisation for faster exec

app.py

@@ -10,33 +10,18 @@ from llm_interaction import get_answer
 vector_database = VectorDatabase() #Instantiate the VectorDatabase Class
 chunks_metadata = []
 
-def main():
-    st.title("Document Query App")
-
-    uploaded_files = st.file_uploader(
-        "Upload PDF or Word files", accept_multiple_files=True, type=["pdf", "docx"]
-    )
 
-    query = st.text_input("Enter your query:")
-
-    if uploaded_files:
-        global chunks_metadata
-        all_chunks, all_embeddings, chunks_metadata = process_files(uploaded_files)
-        vector_database.add_data(all_embeddings, all_chunks, chunks_metadata) # use the method
-
-        st.session_state.files_processed = True
-    
-    if query:
-        results = process_query(query)
-        display_results(results)
 
+@st.cache_data
 def process_query(query):
     if vector_database.is_empty(): #Use the method
         return "Please upload files first."
 
     # query_embedding = get_embeddings([query])[0]
     # results = vector_database.query(query_embedding, k=3) # use the method
+    print('Query:', query)
     query_embedding = get_embeddings([query])[0]  # Get the embedding for the query
+    print('Asking Queries..................')
     results = vector_database.query(query_embedding, k=10)  # Get the top 2 results
 
     return results
@@ -46,17 +31,59 @@ def normalize_line_breaks(text):
 
     return text
 
-def display_results(results):
+def display_results(results, chunks):
     cpt = 1
     for result in (results):
         if result['score'] < 0.5:
-            st.subheader(f"Réponse {cpt+1} :")
-            st.write(f"Source File: {result['file_name']}, Chunk: {result['chunk_index']}, Score: {round((1-result['score'])*100,2)}%")
-            st.subheader("Citations depuis le document :")
-            st.write(normalize_line_breaks(result["chunk_text"]))
-            st_copy_to_clipboard(normalize_line_breaks(result["chunk_text"]))
-            cpt += 1
+            st.subheader(f"Réponse {cpt} :")
+            st.write(f"Source File: {result['file_name']}, Score: {round(1./(1+result['score'])*100,2)}%")     #
+            
+            text_to_display = result['chunk_text']
+            col1, col2 = st.columns(2)
+            with col1:
+                previous_chunk_index = result['chunk_index'] - 1
+                if previous_chunk_index >= 0:
+                    try:
+                        previous_chunk_text = chunks[previous_chunk_index]
+                        if st.button(f"Ajouter la portion de texte précédente", key=f"before_{cpt}"):
+                            text_to_display = previous_chunk_text[:-50] + result['chunk_text']
+                    except IndexError:
+                        pass #silently ignore
+            with col2:
+                next_chunk_index = result['chunk_index'] + 1
+                if next_chunk_index < len(chunks):
+                    try:
+                        next_chunk_text = chunks[next_chunk_index]
+                        if st.button(f"Ajouter la portion de texte suivante", key=f"after_{cpt}"):
+                            text_to_display = result['chunk_text'] + next_chunk_text[50:]
+                    except IndexError:
+                        pass
+
+            st.write("Citations depuis le document :")
+            st.write(normalize_line_breaks(text_to_display))
+            st_copy_to_clipboard(normalize_line_breaks(text_to_display))
+        cpt += 1
+
 
+def main():
+    st.title("Document Query App")
+
+    uploaded_files = st.file_uploader(
+        "Upload PDF or Word files", accept_multiple_files=True, type=["pdf", "docx"]
+    )
+
+    query = st.text_input("Enter your query:")
+
+    if uploaded_files:
+        global chunks_metadata
+        all_chunks, all_embeddings, chunks_metadata = process_files(uploaded_files)
+        vector_database.add_data(all_embeddings, all_chunks, chunks_metadata) # use the method
+
+        st.session_state.files_processed = True
+    
+    if query:
+        results = process_query(query)
+        display_results(results, all_chunks)
 
 if __name__ == "__main__":
     main()

doc_preprocessing.py

@@ -6,8 +6,8 @@ import streamlit as st
 import numpy as np
 import os
 
-emb_model = "intfloat/multilingual-e5-large-instruct"
-emb_model2 = "DeepPavlov/distilrubert-small-cased-conversational"
+# emb_model_ = "intfloat/multilingual-e5-large-instruct"
+emb_model = "intfloat/multilingual-e5-base"
 def extract_text(file):
     text = ""
     # Check if the input is a file path (string) or a file-like object
@@ -69,6 +69,7 @@ def get_embeddings(texts)-> np.ndarray:
         st.error(f"Error generating embeddings: {e}")
         return np.array([])
 
+@st.cache_data
 def process_files(files):
     all_chunks = []
     all_embeddings = []

vector_DB.py

@@ -75,6 +75,7 @@ class VectorDatabase:
         query_embedding = np.array(query_embedding, dtype=np.float32).reshape(1, -1)  # Reshape for FAISS
 
         dist, indices = self.index.search(query_embedding, k=k)
+        
         results = []
         for (i, j) in zip(indices[0], dist[0]):
             chunk_text = self.chunks[i]