diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 93d91cad25..9d267bde6c 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -265,8 +265,8 @@ def LlamaAttention_fast_forward_inference(
         self.paged_attention = torch.empty((2, bsz, n_kv_heads, 2048, head_dim), dtype = dtype, device = "cuda")
         self.paged_attention_K = self.paged_attention[0]
         self.paged_attention_V = self.paged_attention[1]
-        self.paged_attention_K[:,:,:kv_seq_len,:] = Kn
-        self.paged_attention_V[:,:,:kv_seq_len,:] = Vn
+        self.paged_attention_K[:,:,:kv_seq_len,:] = K1
+        self.paged_attention_V[:,:,:kv_seq_len,:] = V1
     pass
     Kn = self.paged_attention_K[:,:,seq_len,:] = K1
     Vn = self.paged_attention_V[:,:,seq_len,:] = V1