diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 296cbf51f5..43f2f1951c 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -664,7 +664,7 @@ def LlamaForCausalLM_fast_forward(
     *args, **kwargs,
 ) -> Union[Tuple, CausalLMOutputWithPast]:
 
-    if causal_mask is None and past_key_value is not None:
+    if causal_mask is None and past_key_values is not None:
         causal_mask = xformers.attn_bias.LowerTriangularMask()
 
     output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions