Should I switch to AWS from Paperspace due to out of memory issues

[ilarum (Muralidharan Surendran)]

Off late paper space seems to running out of memory I am not sure what the issue could be. Refer to the post.

Paperspace running lesson1-rxt50.ipynb, lesson1-vgg results in out of memory issue

I tried to run the lesson1.rxt50 using resnet50 and noticed that when all the layers are unfrozen the resultant execution causes an out of memory exception has anyone encountered this issue and is there a solution. learn.unfreeze() lr=np.array([1e-4,1e-3,1e-2]) learn.fit(lr, 3, cycle_len=1) The last line results in an out of memory. I have noticed that on paperspace I frequently run into the issue of out of memory. I get the below out of memory error cuda runtime error (2) : out of memory at…

cuda runtime error (2) : out of memory at /opt/conda/conda-bld/pytorch_1518244421288/work/torch/lib/THC/generic/THCStorage.cu:58

Should I switch to AWS if its more stable environment

@jeremy @reshama @alessa @rachel

any thoughts on the above.

[kevin_d]

Reducing batch size might be helpful

[alessa (Aless Bandrabur)]

Try different architectures.
Try with smaller - image size and batch size.

I prefer AWS because I am used to it, but also on AWS I have cuda runtime error - specially with resnext50 or resnet101 - but I dont mind since resnet50 gives good enough results.

[ilarum (Muralidharan Surendran)]

@alessa thank you will, will keep your advice. I did contact paperspace and one of their recommendations was to switch to P5000. I might switch over to this, if I keep getting out of memory errors.

[tmcanty (Thomas Canty)]

Is there any comparison on cost? I know after last month trying to train a model ofr the kaggle furniture competition, i got a bit of trouble at home in regards to my paperspace bill :-/