[Bug]: Ideogram4 Failure to load or resume from backups with quantized models

[theotechne]

What happened?

When trying to resume fine-tune training from backup, the program crashes. Probably related to issue #982.

Replacing the unconditional_transformer and text_encoder directories in the backup with those from the original repository resolved the issue, but there should be a code fix.

What did you expect would happen?

The model to load from backup.

Relevant log output

Traceback (most recent call last):
  File "F:\repos\OneTrainer\modules\modelLoader\IdeogramModelLoader.py", line 131, in load
    self.__load_internal(
  File "F:\repos\OneTrainer\modules\modelLoader\IdeogramModelLoader.py", line 38, in __load_internal
    self.__load_diffusers(model, model_type, weight_dtypes, base_model_name, include_unconditional_transformer, quantization)
  File "F:\repos\OneTrainer\modules\modelLoader\IdeogramModelLoader.py", line 74, in __load_diffusers
    text_encoder = self._load_transformers_sub_module(
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "F:\repos\OneTrainer\modules\modelLoader\mixin\HFModelLoaderMixin.py", line 216, in _load_transformers_sub_module
    return self.__load_sub_module(
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "F:\repos\OneTrainer\modules\modelLoader\mixin\HFModelLoaderMixin.py", line 165, in __load_sub_module
    new_value = old_type(value)
                ^^^^^^^^^^^^^^^
  File "F:\repos\OneTrainer\venv\Lib\site-packages\torch\nn\parameter.py", line 57, in __new__
    return torch.Tensor._make_subclass(cls, data, requires_grad)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Only Tensors of floating point and complex dtype can require gradients

Traceback (most recent call last):
  File "F:\repos\OneTrainer\modules\modelLoader\IdeogramModelLoader.py", line 140, in load
    self.__load_diffusers(
  File "F:\repos\OneTrainer\modules\modelLoader\IdeogramModelLoader.py", line 74, in __load_diffusers
    text_encoder = self._load_transformers_sub_module(
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "F:\repos\OneTrainer\modules\modelLoader\mixin\HFModelLoaderMixin.py", line 216, in _load_transformers_sub_module
    return self.__load_sub_module(
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "F:\repos\OneTrainer\modules\modelLoader\mixin\HFModelLoaderMixin.py", line 165, in __load_sub_module
    new_value = old_type(value)
                ^^^^^^^^^^^^^^^
  File "F:\repos\OneTrainer\venv\Lib\site-packages\torch\nn\parameter.py", line 57, in __new__
    return torch.Tensor._make_subclass(cls, data, requires_grad)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Only Tensors of floating point and complex dtype can require gradients

Traceback (most recent call last):
  File "F:\repos\OneTrainer\modules\modelLoader\IdeogramModelLoader.py", line 149, in load
    self.__load_safetensors(
  File "F:\repos\OneTrainer\modules\modelLoader\IdeogramModelLoader.py", line 116, in __load_safetensors
    raise NotImplementedError(
NotImplementedError: Loading single-file safetensors for Ideogram is not supported. Use the diffusers model instead.

Traceback (most recent call last):
  File "F:\repos\OneTrainer\scripts\train.py", line 62, in <module>
    main()
  File "F:\repos\OneTrainer\scripts\train.py", line 49, in main
    trainer.start()
  File "F:\repos\OneTrainer\modules\trainer\GenericTrainer.py", line 129, in start
    self.model = self.model_loader.load(
                 ^^^^^^^^^^^^^^^^^^^^^^^
  File "F:\repos\OneTrainer\modules\modelLoader\GenericFineTuneModelLoader.py", line 53, in load
    base_model_loader.load(model, model_type, model_names, weight_dtypes, quantization)
  File "F:\repos\OneTrainer\modules\modelLoader\IdeogramModelLoader.py", line 158, in load
    raise Exception("could not load model: " + model_names.base_model)

Generate and upload debug_report.log

No response