Update README.md

reword to make more accurate

README.md

@@ -42,7 +42,7 @@ This checkpoint was afterwards finetuned on [tiny_shakespeare](https://huggingfa
 
 I used a modified version of [hivemind's 8bit training script](https://huggingface.co/hivemind/gpt-j-6B-8bit) on 1 Tesla T4 for ~15 minutes
 
-No LORA adapters were used for the sake of easy loading and inference with 🤗. Only Linear biases and LayerNorms were passed to the optimizer.
+No LORA adapters were used for the sake of easy loading and inference with 🤗. Only Linear biases and LayerNorm scales were passed to the optimizer.
 
 End loss: 0.1757839471101761