clarify benefit of prompt loss mask

README.md

@@ -11,7 +11,7 @@ tags: [llama, jupyter, colab, function calling, QLoRa, fine-tuning]
 # *QLoRa Training Notebook for Small Datasets*
 
 Fine-tune HuggingFace language models in Google Colab.
-- Includes a prompt loss-mask for improved performance on small datasets.
+- Includes a prompt loss-mask for improved performance when structured responses are required.
 - Includes a stop token after responses - allowing the model to provide a short reponse (e.g. a function call) and then stop.
 - Allows for fine-tuning 7B and 13B models with the free plan.
 - Allows for fine-tuning 70B models with Google Colab Pro.