GpuArrayException: cuMemAlloc: CUDA_ERROR_OUT_OF_MEMORY: out of memory about text_style_transfer

[fuzhenxin]

Hellos，
Do you change the gpu id of the following line in train.sh:
export THEANO_FLAGS=device=gpu3,floatX=float32,lib.cnmem=0.8

[zhunipingan]

Thanks for your reply!
I noticed the code in train.sh, I think if I use my gpu3, I needn't to do any modification.
But I still have the error, if you know some other reasons? Hhh, I can not solve it by myself now.
Thanks

[fuzhenxin]

Do you modify any code? I'm sure 12G is enough. I did not meet the error ever.

[zhunipingan]

Thanks.
Building f_init... Done Building f_next.. Done Building f_log_probs... Done Building f_cost... Done Computing gradient... Done Building optimizers... Done Optimization Epoch 0 Update 100 style 1 Cost 71.04966 0.34096366 -0.34654176 UD 1.36920809746 Traceback (most recent call last): File "./train_nmt.py", line 55, in <module> 'reload': [False]})
After the 0 epoch, it will say the gpu memeory error. I will try to solve thisn problem. Thank you!

[zhunipingan]

maybe the batch size is too large, I just modify the batch size to 16, and it can be run now.
I have another question, if I want to run the model using another datasets, I think I just need to replace the q_train.txt q_val.txt q_test.txt, r_train.txt r_val.txt r_test.tst,and s_train.txt s_val.txt s_test.tst with my own datasets.
the q_.txt is the same as r_.txt, and the s_.txt is the label of the corresponding q_.txt. I want to know if my understanding is right, thanks~

[fuzhenxin]

Yeah, that is right!

[zhunipingan]

hi,
I met the problem that after I ran the ./test.sh, I found the q_test_style_0.txt was empty and the q_test_style_0.txt was all dots such as '. . . . . . . . . . . . . . . '.
I just modified the batch size in train_nmt.py from 128 to 16.
Can you help me, thanks!!!

[fuzhenxin]

Which model and dataset do you use? I will try to reproduce the srror.

[zhunipingan]

hhh, thank you!
I use your origin dataset, just as you wrote.
$ cd model/style_transfer/session_multi_decoder
$ ./train.sh # train model
$ ./test.sh # test model
$ ./com.sh # show results in compare.txt
And to solve the problem of memory error, I modified the batch size in train_nmy.py to 16.

[fuzhenxin]

I met the same problen........ I am trying to fix it.

[zhunipingan]

hi,
Have you found the solution.

[fuzhenxin]

I can not run it on gpu now due to nvidia cuda or driver error. I am running it on CPU slowly...

[fuzhenxin]

I do not find the solution now..