Update app.py

app.py

@@ -11,6 +11,7 @@ import torch
 import pickle
 import nltk
 import faiss
+import numpy as np
 
 # Ensure NLTK resources are downloaded
 try:
@@ -41,7 +42,7 @@ hf_embeddings = HuggingFaceEmbeddings(model_name='sentence-transformers/all-Mini
 index_path = "faiss_index.pkl"
 if os.path.exists(index_path):
     with open(index_path, "rb") as f:
-        faiss_index = pickle.load(f)
+        faiss_index = faiss.read_index(f)
         print("Loaded FAISS index from faiss_index.pkl")
 else:
     # Create a new FAISS index
@@ -49,6 +50,7 @@ else:
     nlist = 100  # Number of clusters (for IVF)
     quantizer = faiss.IndexFlatL2(d)  # This is the quantizer for IVF
     faiss_index = faiss.IndexIVFFlat(quantizer, d, nlist, faiss.METRIC_L2)
+    faiss_index.train(np.zeros((nlist, d)).astype(np.float32))  # Train the index with dummy data
 
 def extract_text_from_pdf(pdf_path):
     text = ""
@@ -92,7 +94,7 @@ def upload_files(files):
                 embeddings = embedding_model.encode(sentences)
                 
                 for embedding in embeddings:
-                    faiss_index.add(np.array([embedding]))  # Add each embedding individually
+                    faiss_index.add(np.array([embedding]).astype(np.float32))  # Add each embedding individually
 
             except Exception as e:
                 print(f"Error processing file '{file.name}': {e}")
@@ -137,6 +139,3 @@ with gr.Blocks() as demo:
         query_button.click(fn=process_and_query, inputs=[query], outputs=query_output)
 
 demo.launch()
-
-
-