diff --git a/README.md b/README.md
index d2c58e5..eb70f06 100644
--- a/README.md
+++ b/README.md
@@ -193,6 +193,10 @@ print(output.shape)  # Expected shape: (batch_size, target_length, d_model)
 
 ## Advanced Usage
 
+Custom Dataset Training
+
+To train on a custom dataset:
+
 ## Results
 
 My implementation of LongRoPE achieves the following results: