diff --git a/unsloth/__init__.py b/unsloth/__init__.py
index c7f3ea94cf..63689802d9 100644
--- a/unsloth/__init__.py
+++ b/unsloth/__init__.py
@@ -12,5 +12,44 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 __version__ = "2023.11"
+import os
+try:
+	import torch
+except:
+	raise ImportError("Pytorch is not installed. Go to https://pytorch.org/.\n"\
+					  "We have some installation instructions on our Github page.")
+
+# We only support torch 2.1
+major_torch, minor_torch, _ = torch.__version__.split(".")
+major_torch, minor_torch = int(major_torch), int(minor_torch)
+if (major_torch != 2) or (major_torch == 2 and minor_torch < 1):
+	raise ImportError("Unsloth only supports Pytorch 2.1 for now. Please update your Pytorch to 2.1.\n"\
+					  "We have some installation instructions on our Github page.")
+
+# Currently only supports 1 GPU, or else seg faults will occur.
+reload_package = False
+n_gpus = torch.cuda.device_count()
+if n_gpus == 0:
+	raise RuntimeError("Unsloth: Requires at least 1 GPU. Found 0.")
+elif n_gpus > 1:
+	if "CUDA_VISIBLE_DEVICES" in os.environ:
+		device = os.environ["CUDA_VISIBLE_DEVICES"]
+		if not device.isdigit():
+			print(f"Unsloth: 'CUDA_VISIBLE_DEVICES' is currently {device} "\
+				   "but we require 'CUDA_VISIBLE_DEVICES=0'\n"\
+				   "We shall set it ourselves.")
+			os.environ["CUDA_VISIBLE_DEVICES"] = "0"
+			reload_package = True
+	else:
+		print("Unsloth: 'CUDA_VISIBLE_DEVICES' is not set. We shall set it ourselves.")
+		os.environ["CUDA_VISIBLE_DEVICES"] = "0"
+		reload_package = True
+pass
+
+# Reload Pytorch with CUDA_VISIBLE_DEVICES
+if reload_package:
+	import importlib
+	importlib.reload(torch)
+pass
 
 from .models import *
diff --git a/unsloth/models/__init__.py b/unsloth/models/__init__.py
index 46529ccab6..4df2e937e4 100644
--- a/unsloth/models/__init__.py
+++ b/unsloth/models/__init__.py
@@ -12,33 +12,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import torch
-import os
-
-# Currently only supports 1 GPU, or else seg faults will occur.
-reload_package = False
-n_gpus = torch.cuda.device_count()
-if n_gpus == 0:
-	raise RuntimeError("Unsloth: Requires at least 1 GPU. Found 0.")
-elif n_gpus > 1:
-	if "CUDA_VISIBLE_DEVICES" in os.environ:
-		device = os.environ["CUDA_VISIBLE_DEVICES"]
-		if not device.isdigit():
-			print(f"Unsloth: 'CUDA_VISIBLE_DEVICES' is currently {device} "\
-				   "but we require 'CUDA_VISIBLE_DEVICES=0'\n"\
-				   "We shall set it ourselves.")
-			os.environ["CUDA_VISIBLE_DEVICES"] = "0"
-			reload_package = True
-	else:
-		print("Unsloth: 'CUDA_VISIBLE_DEVICES' is not set. We shall set it ourselves.")
-		os.environ["CUDA_VISIBLE_DEVICES"] = "0"
-		reload_package = True
-pass
-
-# Reload Pytorch with CUDA_VISIBLE_DEVICES
-if reload_package:
-	import importlib
-	importlib.reload(torch)
-pass
-
 from .llama import FastLlamaModel