Update app.py

app.py

@@ -48,7 +48,7 @@ class GeminiLLM(Runnable):
 
 class GeminiEmbeddings(Embeddings):
     def __init__(self, model_name="models/embedding-001", api_key=None):
-        api_key = "AIzaSyBIfGJRoet_wzzYXIiWXxStkIigEOzSR2o"
+        self.api_key = api_key or os.environ["GOOGLE_API_KEY"]
         if not api_key:
             raise ValueError("GOOGLE_API_KEY not found in environment variables.")
         os.environ["GOOGLE_API_KEY"] = api_key
@@ -72,7 +72,6 @@ class GeminiEmbeddings(Embeddings):
             task_type="retrieval_query"
         )["embedding"]
 
-reranker = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2')
 
 vectorstore_global = None
 
@@ -82,6 +81,9 @@ if "feedback_log" not in st.session_state:
 
 def load_environment():
     load_dotenv()
+    # Ensure HF_TOKEN is available
+    if "HF_TOKEN" not in os.environ and "HUGGINGFACEHUB_API_TOKEN" in os.environ:
+        os.environ["HF_TOKEN"] = os.environ["HUGGINGFACEHUB_API_TOKEN"]
     genai.configure(api_key=os.environ["GOOGLE_API_KEY"])
 
 def preload_modtran_document():
@@ -195,10 +197,17 @@ def faiss_search_with_keywords(query):
     context= '\n\n'.join([f"[Page {doc.metadata.get('page', 'Unknown')}] {doc.page_content}" for doc in docs])
     return self_reasoning(query, context)
 
+def get_reranker():
+    global reranker
+    if reranker is None:
+        reranker = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2')
+    return reranker
+    
 def faiss_search_with_reasoning(query):
     global vectorstore_global
     if vectorstore_global is None:
         raise ValueError("FAISS vectorstore is not initialized.")
+    reranker = get_reranker()
     retriever = vectorstore_global.as_retriever(search_kwargs={"k": 13})
     docs = retriever.get_relevant_documents(query)
     pairs = [(query, doc.page_content) for doc in docs]