#error out of memory cuda
20 messages · Page 1 of 1 (latest)
Hi, A couple of questions:
- Are you getting the error in your first run or after few runs?
- what is the minimum batch size you tried.
If you are getting this issue after first batch, try to clear the GPU memory with avaiable commands in the pyTorch.
I think you can begin with batch size=1,then try to increase batch size gradually or use gradient accumulation.
I tried batch size 2
And it didn't work
I get it on the first run when the kernel is clean
I tried this one one command pytorch cuda empty cache but didn't work
here's the error: https://pastebin.com/rp0kNXiH
here's my code: https://paste.pythondiscord.com/IVMQ
Pastebin.com is the number one paste tool since 2002. Pastebin is a website where you can store text online for a set period of time.
Restart your kernel completely and reduce the batch size to 1. Also reduce your eval batch size to 1
Monitor the memory usage and increase the batch size accordingly.
isn't it really bad to have a batch size on 1 ?
it's working with bs=1 but is completely frozen
with gpu on 100%
there must be something wrong in the way i coded it
I think your images are too big
Whats the dimensions
Try lowering the dimensions by half and see what happens
i will try thanks
didn't work
nop i can try what does it do ?