diff --git a/train.py b/train.py
index 7433546..9831c2d 100644
--- a/train.py
+++ b/train.py
@@ -45,7 +45,7 @@ wandb_project = 'owt'
 wandb_run_name = 'gpt2' # 'run' + str(time.time())
 # data
 dataset = 'openwebtext'
-gradient_accumulation_steps = 1 # used to simulate larger batch sizes
+gradient_accumulation_steps = 5 # used to simulate larger batch sizes
 batch_size = 12 # if gradient_accumulation_steps > 1, this is the micro-batch size
 block_size = 1024
 # model