diff --git a/ctrl_c_nn.py b/ctrl_c_nn.py
index 6f275dc..67846f1 100644
--- a/ctrl_c_nn.py
+++ b/ctrl_c_nn.py
@@ -4,12 +4,15 @@
 __original_source__ = "https://github.com/manu12121999/ctrl_c_nn/blob/main/ctrl_c_nn.py"
 __email__ = "manu12121999@gmail.com"
 
+import pickle
 import random
+import struct
+import zlib
 import sys
 import math
 import operator
 from multiprocessing import Pool
-import zlib
+from zipfile import ZipFile
 
 sumprod = math.sumprod if sys.version_info >= (3, 12) else lambda p, q: sum([p_i*q_i for p_i, q_i in zip(p, q)])
 
@@ -677,6 +680,60 @@ def forward(self, x: Tensor):
     # missing: weight_init, ConvTranspose2d, MaxPool2d, AvgPool2d, Softmax, BatchNorm2d, InstanceNorm2d, LayerNorm2d, Losses
 
 
+class PthUnpickler(pickle.Unpickler):
+    def __init__(self, picklefile, zipfile):
+        self.zipfile = zipfile
+        super().__init__(picklefile)
+
+    def persistent_load(self, pid):
+        # print("pid", pid)
+        storage, class_type, storage_dir, gpu, size = pid
+        data_list = None
+        with self.zipfile.open(f'best/data/{storage_dir}') as f:
+            data = f.read()
+            if class_type == "int64":
+                data_list = [int.from_bytes(data[i: i+8], "little") for i in range(0, len(data), 8)]
+            elif class_type == "int32":
+                data_list = [int.from_bytes(data[i: i+4], "little") for i in range(0, len(data), 4)]
+            elif class_type == "float":
+                data_list = [struct.unpack('f', data[i: i+4])[0] for i in range(0, len(data), 4)]
+        return data_list
+
+    @staticmethod
+    def load_replacement(data_list, storage_offset, size, stride, *args):
+        assert storage_offset == 0
+        # print("storage:", "--", "size:", size, "stride:", stride)
+        tensor = Tensor(data_list)
+        if len(size) > 1:
+            tensor = tensor.reshape(size)
+        return tensor
+
+    def find_class(self, module, name):
+        # print("m", module, "n", name)
+        if module == 'torch' and name == 'FloatStorage':
+            return "float"
+        elif module == 'torch' and name == 'LongStorage':
+            return "int64"
+        elif module == 'torch' and name == 'IntStorage':
+            return "int32"
+        elif module == 'torch._utils' and name == '_rebuild_tensor_v2':
+            return self.load_replacement  # torch._utils._rebuild_tensor_v2
+        else:
+            print("WARNING: loading module", module, "name", name)
+            return super().find_class(module, name)
+
+
+def load_model(path):
+    with ZipFile(path) as zip_file:
+        with zip_file.open(f'best/data.pkl') as pickle_file:
+            p = PthUnpickler(pickle_file, zip_file)
+            print("loading ...")
+            model_dict = p.load()
+            print("loading completed")
+            print(model_dict["model_state_dict"])
+    return model_dict
+
+
 class ImageIO:
 
     @staticmethod