fix global variable

app.py

@@ -445,6 +445,7 @@ def run_naive_rag_query(query, rag_token_size, prefix, task, few_shot_examples):
     For naive RAG, retrieves top-k chunks (k based on target token size)
     and generates an answer using those chunks.
     """
+    global global_rag_index
     k = max(1, rag_token_size // 256)
     retriever = global_rag_index.as_retriever(search_type="similarity", search_kwargs={"k": k})
     retrieved_docs = retriever.invoke(query)
@@ -464,6 +465,7 @@ def prepare_compression_and_rag(combined_text, retrieval_slider_value, global_lo
     """
     Prepares the compressed KV cache. Uses the precomputed rag_index from state.
     """
+    global global_rag_index
     percentage = int(global_local_value.replace('%', ''))
     question_text = task_description + "\n" + few_shot
     context_encoding = tokenizer(combined_text, return_tensors="pt").to(device)