diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 1d413f0acc..68aa527990 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -1706,18 +1706,19 @@ class FastLlamaModel:
         spaces = re.search('\n([\s\t]{1,})', original_debug).group(0)[1:]
         front_spaces = re.match('([\s\t]{1,})', inner_training_loop).group(0)
 
-        debug_info = """debug_info = \\
         unsloth_0 = r'==((====))=='
         unsloth_1 = r'   \\   /|  '
         unsloth_2 = r'O^O/ \_/ \  '
         unsloth_3 = r'\        /  '
         unsloth_4 = r' "-____-"   '
+
+        debug_info = """debug_info = \\
         f"{unsloth_0}  Unsloth - 2x faster free finetuning | Num GPUs = {args.world_size}\\n"\\
         f"{unsloth_1}  Num examples = {num_examples:,} | Num Epochs = {num_train_epochs:,}\\n"\\
         f"{unsloth_2}  Batch size per device = {self._train_batch_size:,} | Gradient Accumulation steps = {args.gradient_accumulation_steps}\\n"\\
         f"{unsloth_3}  Total batch size = {total_train_batch_size:,} | Total steps = {max_steps:,}\\n"\\
         f'{unsloth_4}  Number of trainable parameters = {get_model_param_count(model, trainable_only=True):,}'
-        print(debug_info)
+        logger.warning(debug_info)
         import subprocess, re, gc, numpy as np
         a = np.array([0,])
         try:
@@ -1733,6 +1734,14 @@ class FastLlamaModel:
             gc.collect()
             torch.cuda.empty_cache()"""
 
+        print(debug_info)
+        debug_info = debug_info\
+            .replace("{unsloth_0}", unsloth_0)\
+            .replace("{unsloth_1}", unsloth_1)\
+            .replace("{unsloth_2}", unsloth_2)\
+            .replace("{unsloth_3}", unsloth_3)\
+            .replace("{unsloth_4}", unsloth_4)
+
         debug_info = debug_info.split('\n')
         debug_info = "\n".join([debug_info[0]] + [spaces + x[8:] for x in debug_info[1:]])
         inner_training_loop = inner_training_loop.replace(original_debug, debug_info)