diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index fa3c8d1c5a..052ff54282 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -750,7 +750,7 @@ def LlamaForCausalLM_fast_forward(
     # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
     self.model._has_no_labels = labels is None
 
-    if past_key_value is not None and \
+    if past_key_values is not None and \
         hasattr(self.model.layers[0].self_attn, "paged_attention"):
         outputs = LlamaModel_fast_forward_inference(
             self.model,
diff --git a/unsloth/models/mistral.py b/unsloth/models/mistral.py
index 99dc3c1fd9..cd92d0e7dd 100644
--- a/unsloth/models/mistral.py
+++ b/unsloth/models/mistral.py
@@ -201,7 +201,7 @@ def MistralForCausalLM_fast_forward(
     # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
     self.model._has_no_labels = labels is None
 
-    if past_key_value is not None and \
+    if past_key_values is not None and \
         hasattr(self.model.layers[0].self_attn, "paged_attention"):
         outputs = LlamaModel_fast_forward_inference(
             self.model,