CUDA out of memory with 'ruDALLE-image-prompts-A100.ipynb' & 'ruDALLE-image-prompts-dress-mannequins-V100.ipynb'

[Limbicnation]

Hi I get an OOM error with an (RTX 3060 12 GB) for the following scripts: 'uDALLE-image-prompts-A100.ipynb' , 'ruDALLE-image-prompts-dress-mannequins-V100.ipynb'

Unfortunately I could not make it work. I tried to look at the other script 'Malevich-3.5GB-vRAM-usage.ipynb' and I have tried to adept the code, but I still get an error:

RuntimeError: CUDA out of memory. Tried to allocate 1.98 GiB (GPU 0; 11.77 GiB total capacity; 6.63 GiB already allocated; 1.08 GiB free; 11.00 GiB allowed; 7.77 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Any help on how to tweak the code would be appreciated!

Thanks! Gero

[PavelValentov]

The same issue:

Traceback (most recent call last):
  File "D:\code\ru-dalle\py\test.py", line 17, in <module>
    dalle = get_rudalle_model('Malevich', pretrained=True, fp16=True, device=device)
  File "C:\Python310\lib\site-packages\rudalle\dalle\__init__.py", line 148, in get_rudalle_model
    model = model.to(device)
  File "C:\Python310\lib\site-packages\rudalle\dalle\fp16.py", line 63, in to
    self.module.to(device)
  File "C:\Python310\lib\site-packages\rudalle\dalle\model.py", line 165, in to
    return super().to(device, *args, **kwargs)
  File "C:\Python310\lib\site-packages\torch\nn\modules\module.py", line 907, in to
    return self._apply(convert)
  File "C:\Python310\lib\site-packages\torch\nn\modules\module.py", line 578, in _apply
    module._apply(fn)
  File "C:\Python310\lib\site-packages\torch\nn\modules\module.py", line 578, in _apply
    module._apply(fn)
  File "C:\Python310\lib\site-packages\torch\nn\modules\module.py", line 578, in _apply
    module._apply(fn)
  [Previous line repeated 2 more times]
  File "C:\Python310\lib\site-packages\torch\nn\modules\module.py", line 601, in _apply
    param_applied = fn(param)
  File "C:\Python310\lib\site-packages\torch\nn\modules\module.py", line 905, in convert
    return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking)
RuntimeError: CUDA out of memory. Tried to allocate 24.00 MiB (GPU 0; 2.00 GiB total capacity; 1.69 GiB already allocated; 0 bytes free; 1.72 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
P

[Pipazoul]

You will at least need 40go of vram, the model itself loads 20go into vram. Maybe use a less ressource intensive model Malevich for ex : https://colab.research.google.com/drive/1AoolDYePUpPkRCKIu0cP9zV7lX5QGD3Z?usp=sharing