diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index f55d71c188..747ef0793a 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -306,6 +306,7 @@ def LlamaDecoderLayer_fast_forward(
     bsz, q_len, hd = hidden_states.size()
 
     if (not self.training and q_len == 1):
+        print(1)
         # Self Attention
         residual = hidden_states
         hidden_states = fast_rms_layernorm_inference(self.input_layernorm, hidden_states)