lusxvr/nanoVLM-222M image patching issue

running: python generate.py --hf_model lusxvr/nanoVLM-222M

python generate.py --hf_model lusxvr/nanoVLM-222M
Using device: cuda
Loading weights from: lusxvr/nanoVLM-222M
Traceback (most recent call last):
File "D:\_src\nanoVLM\generate.py", line 71, in
main()
File "D:\_src\nanoVLM\generate.py", line 55, in main
processed_image, splittedimage_count = image_processor(img)
^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gilad\AppData\Local\Programs\Python\Python311\Lib\site-packages\torchvision\transforms\transforms.py", line 95, in __call__
img = t(img)
^^^^^^
File "C:\Users\gilad\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\gilad\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\_src\nanoVLM\data\custom_transforms.py", line 98, in forward
raise ValueError(f'Image size {(h,w)} not divisible by patch_size {self.p}')
ValueError: Image size (1024, 896) not divisible by patch_size 224

its seems that the dynamic resize the image to (1024, 896), but than the SplitImage cant handle this size,

help anyone

Run `python generate.py --hf_model lusxvr/nanoVLM-222M` to reproduce the failure, then inspect `data/custom_transforms.py` at line 98 and the call from `generate.py` line 55. Determine how the 1024x896 image reaches the patching transform; done means this input no longer raises the reported divisibility error.

Written by the indexing model from the issue text.