app.py for consuming hsmw serialized model

app.py

@@ -1,16 +1,17 @@
-
 import streamlit as st
 import joblib
 import numpy as np
-import faiss
+import hnswlib
 import os
 from openai import OpenAI
 
-# Initialize OpenAI client using custom Hugging Face secret
+# Initialize OpenAI client using secret from Hugging Face Spaces
 client = OpenAI(api_key=os.getenv("POCJujitsu"))
 
-# Load serialized FAISS index and document chunks
-chunks, index = joblib.load("rag_model.joblib")
+# Load serialized HNSW index and document chunks
+model_data = joblib.load("rag_model_hnsw.joblib")
+chunks = model_data["chunks"]
+index = model_data["index"]
 
 # Embed query using OpenAI embedding API
 def embed_query(text):
@@ -20,23 +21,13 @@ def embed_query(text):
     )
     return np.array(response.data[0].embedding, dtype=np.float32).reshape(1, -1)
 
-# Semantic search using FAISS (for older FAISS versions)
-
-# Semantic search with fallback handling
-
-# Semantic search using FAISS - strictly for older API with preallocated arrays
+# Semantic search using HNSWlib
 def search(query, k=3):
     query_vec = embed_query(query).astype(np.float32)
-
-    # Preallocate arrays (required for FAISS IndexFlatL2 in older versions)
-    distances = np.empty((1, k), dtype=np.float32)
-    labels = np.empty((1, k), dtype=np.int64)
-
-    # Call FAISS with all required arguments
-    index.search(query_vec, k, distances, labels)
-
+    labels, distances = index.knn_query(query_vec, k=k)
     return [chunks[i] for i in labels[0]]
 
+# Chat modes
 def chat_no_rag(question):
     response = client.chat.completions.create(
         model="gpt-3.5-turbo",
@@ -48,8 +39,12 @@ def chat_no_rag(question):
 
 def chat_with_rag(question, context_chunks):
     context = "\n".join(context_chunks)
-    prompt = f"Ayudate en el siguiente contexto para responder la pregunta y usa tus conocimientos en caso de ser necesario: \n\n{context}\n\nPregunta: {question}"
-
+    prompt = (
+        "Usa el siguiente contexto como referencia para responder la pregunta. "
+        "Puedes complementar con tus propios conocimientos si es necesario.\n\n"
+        f"Contexto:\n{context}\n\n"
+        f"Pregunta: {question}\nRespuesta:"
+    )
     response = client.chat.completions.create(
         model="gpt-3.5-turbo",
         messages=[{"role": "user", "content": prompt}],