diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index c4cdc5b82e..82b80e2dc9 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -992,7 +992,7 @@ class FastLlamaModel:
         logger.warning_once(debug_info)"""
 
         debug_info = debug_info.split('\n')
-        debug_info = "\n".join([debug_info[0]] + [spaces + x.lstrip("\t ") for x in debug_info[1:]])
+        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)
 
         debug_info = """n_total_devices = total_train_batch_size // \\
@@ -1006,7 +1006,7 @@ class FastLlamaModel:
             )
         debug_info ="""
         debug_info = debug_info.split('\n')
-        debug_info = "\n".join([debug_info[0]] + [spaces + x.lstrip("\t ") for x in debug_info[1:]])
+        debug_info = "\n".join([debug_info[0]] + [spaces + x[8:] for x in debug_info[1:]])
         inner_training_loop = inner_training_loop.replace("debug_info =", debug_info, 1)
 
         front_spaces = re.match(r"[\t\s]{1,}", inner_training_loop).group(0)
@@ -1039,7 +1039,7 @@ class FastLlamaModel:
                 divisor = n_total_devices / 2 / 2
                 ga = args.gradient_accumulation_steps = max(int(ga / divisor), 1)"""
         check_batches = check_batches.split('\n')
-        check_batches = "\n".join([check_batches[0]] + [front_spaces + x.lstrip("\t ") for x in check_batches[1:]])
+        check_batches = "\n".join([check_batches[0]] + [front_spaces + x[8:] for x in check_batches[1:]])
         inner_training_loop = inner_training_loop.replace(
             "train_dataloader = self.get_train_dataloader()",
             check_batches, 1,
diff --git a/unsloth/models/mistral.py b/unsloth/models/mistral.py
index 10362a7885..ad119af146 100644
--- a/unsloth/models/mistral.py
+++ b/unsloth/models/mistral.py
@@ -397,7 +397,7 @@ class FastMistralModel(FastLlamaModel):
         logger.warning_once(debug_info)"""
 
         debug_info = debug_info.split('\n')
-        debug_info = "\n".join([debug_info[0]] + [spaces + x.lstrip("\t ") for x in debug_info[1:]])
+        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)
 
         debug_info = """n_total_devices = total_train_batch_size // \\
@@ -411,7 +411,7 @@ class FastMistralModel(FastLlamaModel):
             )
         debug_info ="""
         debug_info = debug_info.split('\n')
-        debug_info = "\n".join([debug_info[0]] + [spaces + x.lstrip("\t ") for x in debug_info[1:]])
+        debug_info = "\n".join([debug_info[0]] + [spaces + x[8:] for x in debug_info[1:]])
         inner_training_loop = inner_training_loop.replace("debug_info =", debug_info, 1)
 
         front_spaces = re.match(r"[\t\s]{1,}", inner_training_loop).group(0)
@@ -444,7 +444,7 @@ class FastMistralModel(FastLlamaModel):
                 divisor = n_total_devices / 2 / 2
                 ga = args.gradient_accumulation_steps = max(int(ga / divisor), 1)"""
         check_batches = check_batches.split('\n')
-        check_batches = "\n".join([check_batches[0]] + [front_spaces + x.lstrip("\t ") for x in check_batches[1:]])
+        check_batches = "\n".join([check_batches[0]] + [front_spaces + x[8:] for x in check_batches[1:]])
         inner_training_loop = inner_training_loop.replace(
             "train_dataloader = self.get_train_dataloader()",
             check_batches, 1,