diff --git a/pyproject.toml b/pyproject.toml
index 1d206913a9..01636e75f1 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -40,7 +40,7 @@ triton = [
 ]
 
 huggingface = [
-    "unsloth_zoo>=2025.3.4",
+    "unsloth_zoo>=2025.3.5",
     "packaging",
     "tyro",
     "transformers>=4.46.1,!=4.47.0",
@@ -354,7 +354,7 @@ colab-ampere-torch220 = [
     "flash-attn>=2.6.3",
 ]
 colab-new = [
-    "unsloth_zoo>=2025.3.4",
+    "unsloth_zoo>=2025.3.5",
     "packaging",
     "tyro",
     "transformers>=4.46.1,!=4.47.0",
diff --git a/unsloth/__init__.py b/unsloth/__init__.py
index 4336ec494b..9ed356db54 100644
--- a/unsloth/__init__.py
+++ b/unsloth/__init__.py
@@ -198,7 +198,7 @@ pass
 # Check for unsloth_zoo
 try:
     unsloth_zoo_version = importlib_version("unsloth_zoo")
-    if Version(unsloth_zoo_version) < Version("2025.3.4"):
+    if Version(unsloth_zoo_version) < Version("2025.3.5"):
         try:
             os.system("pip install --upgrade --no-cache-dir --no-deps unsloth_zoo")
         except:
diff --git a/unsloth/models/_utils.py b/unsloth/models/_utils.py
index c01e0ccc82..7ac35d71b4 100644
--- a/unsloth/models/_utils.py
+++ b/unsloth/models/_utils.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-__version__ = "2025.3.6"
+__version__ = "2025.3.7"
 
 __all__ = [
     "SUPPORTS_BFLOAT16",
@@ -1052,8 +1052,7 @@ def _unsloth_pre_compute_loss(self, model, inputs, *args, **kwargs):
 
     # Get gradient accumulation steps if possible
     if num_items_in_batch is None and \
-        getattr(self, "args", {}).get("gradient_accumulation_steps", 1) != 1:
-
+        getattr(getattr(self, "args", self), "gradient_accumulation_steps", 1) != 1:
         name = (model.base_model.model if hasattr(model, "base_model") else model).__class__.__name__
         logger.warning_once(
             f"Unsloth: Not an error, but {name} does not accept `num_items_in_batch`.\n"\
diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 3dacf5cdd5..a490fb8ab4 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -1843,7 +1843,7 @@ class FastLlamaModel:
             else:
                 inner_training_loop = Trainer._original_training_loop
         except:
-            raise RuntimeError('Unsloth currently does not support multi GPU setups - but we are working on it!')
+            raise RuntimeError('Unsloth: Unsuccessfully patched inner_training_loop')
         pass
         
         import transformers.trainer
@@ -1869,7 +1869,7 @@ class FastLlamaModel:
         f"{chr(92)}        /    Data Parallel GPUs = {args.world_size} | Total batch size ({self._train_batch_size} x {args.gradient_accumulation_steps} x {args.world_size}) = {total_train_batch_size:,}\\n"\\
         f' "-____-"     Trainable parameters = {get_model_param_count(model, trainable_only=True):,}/{get_model_param_count(model):,} ({get_model_param_count(model, trainable_only=True)/get_model_param_count(model)*100:.2f}% trained)'
         logger.warning(debug_info)
-        import subprocess, re, gc
+        import gc
         for _ in range(3):
             gc.collect()
             torch.cuda.empty_cache()"""
@@ -1897,9 +1897,6 @@ class FastLlamaModel:
             "_inner_training_loop",
             "_fast_inner_training_loop", 1,
         )
-        exec(inner_training_loop, globals())
-
-        Trainer._inner_training_loop = _fast_inner_training_loop
         inner_training_loop = inner_training_loop.replace(
             "is_torch_tpu_available()",
             "False",