diff --git a/library/adafactor_fused.py b/library/adafactor_fused.py
index 6cf1ac1a..47a4e374 100644
--- a/library/adafactor_fused.py
+++ b/library/adafactor_fused.py
@@ -30,7 +30,7 @@ def copy_stochastic_(target: torch.Tensor, source: torch.Tensor):
 
 # Kahan summation for bfloat16
 # The implementation was provided by araleza.
-# Base on paper "Revisiting BFloat16 Training": https://arxiv.org/pdf/2010.06192
+# Based on paper "Revisiting BFloat16 Training": https://arxiv.org/pdf/2010.06192
 
 kahan_residuals = []
 tensor_index = 0