fix for itemsize => element_size() for torch backwards compat (#30133)

* fix for itemsize => element_size() for torch backwards compat

* improve handling of element counting

* Update src/transformers/modeling_utils.py

* fixup

* Update src/transformers/modeling_utils.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

---------
Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>
Co-authored-by: Younes Belkada <younesbelkada@gmail.com>
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

src/transformers/modeling_utils.py

@@ -1160,12 +1160,13 @@ class ModuleUtilsMixin:
                 # For 4bit models, we need to multiply the number of parameters by 2 as half of the parameters are
                 # used for the 4bit quantization (uint8 tensors are stored)
                 if is_loaded_in_4bit and isinstance(param, bnb.nn.Params4bit):
-                    quant_storage = self.hf_quantizer.quantization_config.bnb_4bit_quant_storage
-                    # For compatibility with older PT version - see: https://github.com/huggingface/peft/pull/1635
-                    nb_params = (
-                        quant_storage.itemsize if hasattr(quant_storage, "itemsize") else quant_storage.element_size()
-                    )
-                    total_numel.append(param.numel() * 2 * nb_params)
+                    if hasattr(param, "element_size"):
+                        num_bytes = param.element_size()
+                    elif hasattr(param, "quant_storage"):
+                        num_bytes = param.quant_storage.itemsize
+                    else:
+                        num_bytes = 1
+                    total_numel.append(param.numel() * 2 * num_bytes)
                 else:
                     total_numel.append(param.numel())