# Outofmemory error when running pipieline.to("cuda") **URL:** \ **Category:** 🧨 Diffusers\ **Created:** [March 16, 2023, 4:08pm UTC](https://discuss.huggingface.co/t/outofmemory-error-when-running-pipieline-to-cuda/33956 "2023-03-16T16:08:29Z")\ **Posts on this page:** 2\ **Page:** 1
**Author:** ![winecoding](https://avatars.discourse-cdn.com/v4/letter/w/839c29/32.png) [@winecoding](https://discuss.huggingface.co/u/winecoding)\ **Post date:** [March 16, 2023, 4:08pm UTC](https://discuss.huggingface.co/t/outofmemory-error-when-running-pipieline-to-cuda/33956/1 "2023-03-16T16:08:29Z")
I am trying to reproducing the [quickguide of diffuser](https://github.com/huggingface/diffusers), and got the following error messages when running **pipeline.to(“cuda”)** Which part in pipeline class I should modify to solve this issue? > OutOfMemoryError Traceback (most recent call last) > Cell In[4], line 1 > ----\> 1 pipeline.to(“cuda”) > > File C:\ProgramData\Miniconda3\envs\hg\_diffuser\lib\site-packages\diffusers\pipelines\pipeline\_utils.py:396, in DiffusionPipeline.to(self, torch\_device, silence\_dtype\_warnings) > 383 if ( > 384 module.dtype == torch.float16 > 385 and str(torch\_device) in [“cpu”] > 386 and not silence\_dtype\_warnings > 387 and not is\_offloaded > 388 ): > 389 logger.warning( > 390 “Pipelines loaded with `torch_dtype=torch.float16` cannot run with `cpu` device. It” > 391 " is not recommended to move them to `cpu` as running them will fail. Please make" > (…) > 394 " `torch_dtype=torch.float16` argument, or use another device for inference." > 395 ) > → 396 module.to(torch\_device) > 397 return self > > File C:\ProgramData\Miniconda3\envs\hg\_diffuser\lib\site-packages\torch\nn\modules\module.py:1145, in Module.to(self, \*args, \*\*kwargs) > 1141 return t.to(device, dtype if t.is\_floating\_point() or t.is\_complex() else None, > 1142 non\_blocking, memory\_format=convert\_to\_format) > 1143 return t.to(device, dtype if t.is\_floating\_point() or t.is\_complex() else None, non\_blocking) > → 1145 return self.\_apply(convert) > > File C:\ProgramData\Miniconda3\envs\hg\_diffuser\lib\site-packages\torch\nn\modules\module.py:797, in Module.\_apply(self, fn) > 795 def \_apply(self, fn): > 796 for module in self.children(): > → 797 module.\_apply(fn) > 799 def compute\_should\_use\_set\_data(tensor, tensor\_applied): > 800 if torch.\_has\_compatible\_shallow\_copy\_type(tensor, tensor\_applied): > 801 # If the new tensor has compatible tensor type as the existing tensor, > 802 # the current behavior is to change the tensor in-place using `.data =`, > (…) > 807 # global flag to let the user control whether they want the future > 808 # behavior of overwriting the existing tensor or not. > > File C:\ProgramData\Miniconda3\envs\hg\_diffuser\lib\site-packages\torch\nn\modules\module.py:797, in Module.\_apply(self, fn) > 795 def \_apply(self, fn): > 796 for module in self.children(): > → 797 module.\_apply(fn) > 799 def compute\_should\_use\_set\_data(tensor, tensor\_applied): > 800 if torch.\_has\_compatible\_shallow\_copy\_type(tensor, tensor\_applied): > 801 # If the new tensor has compatible tensor type as the existing tensor, > 802 # the current behavior is to change the tensor in-place using `.data =`, > (…) > 807 # global flag to let the user control whether they want the future > 808 # behavior of overwriting the existing tensor or not. > > ``` > [... skipping similar frames: Module._apply at line 797 (2 times)] > > ``` > > File C:\ProgramData\Miniconda3\envs\hg\_diffuser\lib\site-packages\torch\nn\modules\module.py:797, in Module.\_apply(self, fn) > 795 def \_apply(self, fn): > 796 for module in self.children(): > → 797 module.\_apply(fn) > 799 def compute\_should\_use\_set\_data(tensor, tensor\_applied): > 800 if torch.\_has\_compatible\_shallow\_copy\_type(tensor, tensor\_applied): > 801 # If the new tensor has compatible tensor type as the existing tensor, > 802 # the current behavior is to change the tensor in-place using `.data =`, > (…) > 807 # global flag to let the user control whether they want the future > 808 # behavior of overwriting the existing tensor or not. > > File C:\ProgramData\Miniconda3\envs\hg\_diffuser\lib\site-packages\torch\nn\modules\module.py:820, in Module.\_apply(self, fn) > 816 # Tensors stored in modules are graph leaves, and we don’t want to > 817 # track autograd history of `param_applied`, so we have to use > 818 # `with torch.no_grad():` > 819 with torch.no\_grad(): > → 820 param\_applied = fn(param) > 821 should\_use\_set\_data = compute\_should\_use\_set\_data(param, param\_applied) > 822 if should\_use\_set\_data: > > File C:\ProgramData\Miniconda3\envs\hg\_diffuser\lib\site-packages\torch\nn\modules\module.py:1143, in Module.to..convert(t) > 1140 if convert\_to\_format is not None and t.dim() in (4, 5): > 1141 return t.to(device, dtype if t.is\_floating\_point() or t.is\_complex() else None, > 1142 non\_blocking, memory\_format=convert\_to\_format) > → 1143 return t.to(device, dtype if t.is\_floating\_point() or t.is\_complex() else None, non\_blocking) > > OutOfMemoryError: CUDA out of memory. Tried to allocate 58.00 MiB (GPU 0; 4.00 GiB total capacity; 3.41 GiB already allocated; 0 bytes free; 3.47 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 ---
**Author:** ![williamberman](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/williamberman/32/14112_2.png) [@williamberman](https://discuss.huggingface.co/u/williamberman)\ **Post date:** [March 16, 2023, 11:51pm UTC](https://discuss.huggingface.co/t/outofmemory-error-when-running-pipieline-to-cuda/33956/2 "2023-03-16T23:51:57Z")
Hey @winecoding it looks like your gpu only has 4GB of memory. We support a few memory optimizations documented here [Memory and speed](https://huggingface.co/docs/diffusers/optimization/fp16) I would recommend looking into accelerate offloading