diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 6abd49222..48628a8d0 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -1787,6 +1787,11 @@ def unsloth_fast_generate(
 
 class FastLlamaModel:
 
+    @staticmethod
+    def _prepare_for_qat(model, qat_scheme):
+        model = _prepare_model_for_qat(model, qat_scheme)
+        return model
+
     @staticmethod
     def pre_patch():
         init_name, function = patch_llama_rope_scaling(
@@ -2668,7 +2673,8 @@ def get_peft_model(
         # Apply QAT + LoRA if specified
         if qat_scheme is not None:
             print("Unsloth: Applying QAT to mitigate quantization degradation")
-            model = _prepare_model_for_qat(model, qat_scheme)
+            model = FastLlamaModel._prepare_for_qat(model, qat_scheme)
+            
         pass
 
         model._saved_temp_tokenizer = _saved_temp_tokenizer
diff --git a/unsloth/models/loader.py b/unsloth/models/loader.py
index 5f92999f4..f31d82106 100644
--- a/unsloth/models/loader.py
+++ b/unsloth/models/loader.py
@@ -569,6 +569,12 @@ def from_pretrained(
 
 
 class FastModel(FastBaseModel):
+
+    @staticmethod
+    def _prepare_for_qat(model, qat_scheme):
+        model = _prepare_model_for_qat(model, qat_scheme)
+        return model
+
     @staticmethod
     def from_pretrained(
         model_name                 = "unsloth/Llama-3.2-11B-Vision-Instruct-bnb-4bit",
@@ -1070,7 +1076,7 @@ def from_pretrained(
         # Apply QAT if specified
         if qat_scheme is not None:
             print("Unsloth: Applying QAT to mitigate quantization degradation")
-            model = _prepare_model_for_qat(model, qat_scheme)
+            model = FastModel._prepare_for_qat(model, qat_scheme)
         pass
 
         return model, tokenizer