diff --git a/app/models.py b/app/models.py
index bf27efc..b426cb2 100644
--- a/app/models.py
+++ b/app/models.py
@@ -1,7 +1,7 @@
 from sentence_transformers import SentenceTransformer, CrossEncoder
 from transformers import pipeline
 import torch
-
+import nltk
 
 bi_encoder = SentenceTransformer('multi-qa-MiniLM-L6-cos-v1')
 
@@ -9,3 +9,5 @@
 cross_encoder_large = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2')
 
 qa_model = pipeline('question-answering', model='deepset/roberta-base-squad2')
+
+nltk.download('punkt')
diff --git a/app/search_logic.py b/app/search_logic.py
index 34673ea..b1595c3 100644
--- a/app/search_logic.py
+++ b/app/search_logic.py
@@ -25,7 +25,6 @@
 BI_ENCODER_CANDIDATES = 60 if torch.cuda.is_available() else 20
 SMALL_CROSS_ENCODER_CANDIDATES = 30 if torch.cuda.is_available() else 10
 
-nltk.download('punkt')
 logger = logging.getLogger(__name__)