fix typo

src/axolotl/utils/models.py

@@ -333,7 +333,7 @@ def load_model(
             model, use_gradient_checkpointing=cfg.gradient_checkpointing
         )
 
-        # LlamaRMSNorm layers are in fp32 after kit call, so we need to
+        # LlamaRMSNorm layers are in fp32 after kbit_training, so we need to
         # convert them back to fp16/bf16 for flash-attn compatibility.
         if cfg.flash_attention and cfg.is_llama_derived_model:
             for name, module in model.named_modules():