Hi to all,
Is it possible to upload any external function or class present in a external module on GPU in pytorch?
I tried to do it by
but it says ::
AttributeError: ‘list’ object has no attribute ‘to’
Hi to all,
No, you cannot move arbitrary objects to the GPU as the underlying library/frameworks needs to support GPUs. Your current code snippet fails since you are trying to move a plain
list to the GPU, which Python doesn’t understand as it does not support GPUs natively.
Instead, create a PyTorch tensor and move this object to the device.
for i in range(m):
for j in range(m):
for k in range(m):
for l in range(m):
for p in range(n):
for i in Mu:
This is the function i am talking about. in short it is giving a nested lists as [[1,1,1,1,1][2,1,2,0,1]…]. It is a tensor, if i send it .to(dev). Will it be on GPU?
Your code is quite unreadable as it’s not properly formatted, but I assume you want to call
to(device) on the returned tensor. If so, then yes the result will be moved the the GPU.
I have a question, while running the code on GPU, my code is finished running but the memory allocated to the GPU is still remain. I want to remove this memory. I tried del tensors, model etc. but it is still there.
I used command nvtop to see the usage and stats of GPU, after finishing it shows this, 894MiB GPU memory is still there. How can I completely remove it.
gauge_neural.py is the name of my python script.
PID USER DEV TYPE GPU GPU MEM CPU HOST MEM Command
1265975 vchahar 0 Compute 0% 894MiB 8% 0% 2429MiB python3 gauge_neural.py
The ~900MB could be used by the CUDA context and will be freed once you close your application.