Update app.py

app.py

@@ -7,6 +7,9 @@ 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)
@@ -18,7 +21,6 @@ def extract_text_from_pdf(pdf_path):
 
 # 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])
@@ -27,19 +29,16 @@ def build_faiss_index(documents):
     faiss.write_index(index, "apexcustoms_index.faiss")
     model.save("sentence_transformer_model")
 
-    return index, model
+    return index
 
 # 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, model = build_faiss_index(documents)
+    index = 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")
@@ -50,8 +49,12 @@ def retrieve_documents(query, k=5):
     return [documents[i] for i in indices[0]]
 
 def respond(message, history, system_message, max_tokens, temperature, top_p):
+    # Retrieve relevant documents
     relevant_docs = retrieve_documents(message)
-    context = "\n\n".join(relevant_docs)
+    context = "\n\n".join(relevant_docs[:3])  # Limit context to top 3 documents
+
+    # Limit history to the last 5 exchanges to reduce payload size
+    history = history[-5:]
 
     messages = [{"role": "system", "content": system_message},
                 {"role": "user", "content": f"Context: {context}\n\n{message}"}]