Update app.py

app.py

@@ -7,50 +7,41 @@ import numpy as np
 from huggingface_hub import InferenceClient
 from sentence_transformers import SentenceTransformer
 
-# Initialize the SentenceTransformer model
-model = SentenceTransformer('paraphrase-MiniLM-L6-v2')
+
+
 
 # Extract text from PDF
 def extract_text_from_pdf(pdf_path):
     doc = fitz.open(pdf_path)
-    text = ""
-    for page_num in range(doc.page_count):
-        page = doc.load_page(page_num)
-        text += page.get_text()
-    return text.split("\n\n")
 
 # Build FAISS index
 def build_faiss_index(documents):
+    model = SentenceTransformer('paraphrase-MiniLM-L6-v2')
     document_embeddings = model.encode(documents)
 
     index = faiss.IndexFlatL2(document_embeddings.shape[1])
-    index.add(document_embeddings)
-
     faiss.write_index(index, "apexcustoms_index.faiss")
     model.save("sentence_transformer_model")
 
-    return index
+    return index, model
 
 # Ensure that text extraction and FAISS index building is done
 if not os.path.exists("apexcustoms_index.faiss") or not os.path.exists("sentence_transformer_model"):
     documents = extract_text_from_pdf("apexcustoms.pdf")
     with open("apexcustoms.json", "w") as f:
         json.dump(documents, f)
-    index = build_faiss_index(documents)
+    index, model = build_faiss_index(documents)
 else:
     index = faiss.read_index("apexcustoms_index.faiss")
+    model = SentenceTransformer('sentence_transformer_model')
+    with open("apexcustoms.json", "r") as f:
+        documents = json.load(f)
 
 # Hugging Face client
 client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
-
-def retrieve_documents(query, k=5):
-    query_embedding = model.encode([query])
-    distances, indices = index.search(query_embedding, k)
     return [documents[i] for i in indices[0]]
 
-async def respond(message, history, system_message, max_tokens, temperature, top_p):
-    # Retrieve relevant documents
-    relevant_docs = retrieve_documents(message)
+def respond(message, history, system_message, max_tokens, temperature, top_p):
     context = "\n\n".join(relevant_docs[:3])  # Limit context to top 3 documents
 
     # Limit history to the last 5 exchanges to reduce payload size