lstm-char-cnn-tensorflow icon indicating copy to clipboard operation
lstm-char-cnn-tensorflow copied to clipboard

Published 20 hours ago • verified publisher icon carpedm20

dataset : nsmc - ResourceExhaustedError ?

Open jiwoongim opened this issue 8 years ago • 2 comments

I have a question about running on nsmc dataset python main.py --dataset nsmc doesn't seem to work on both gtx1080 (8 GB) and k40 (12 GB). I get such an error " ... tensorflow.python.framework.errors.ResourceExhaustedError: OOM when allocating tensor with shape[100,450537] ..." Is this "python main.py --dataset nsmc" supposed to run on a specific gpu? are there certain requirements?

jiwoongim avatar Jul 14 '16 18:07 jiwoongim

No, I tested this code on gtx980ti so 1080 should fit well. But I've never tried nsmc before because ptb still couldn't converge well as I discussed in #3. Was it okay to run ptb instead of nsmc?

carpedm20 avatar Jul 14 '16 18:07 carpedm20

Python main.py doesn't crush and runs on GTX1080. However it fails to train at all. When I trained on k40, python main.py works. I get "{'perplexity': 88.218780156903421, 'epoch': 24, 'valid_perplexity': 150.16004922594567, 'learning_rate': 0.0001220703125}

Test: loss: 4.916528, perplexity: 136.527725 [*] Test loss: 4.972944, perplexity: 144.451502"

jiwoongim avatar Jul 19 '16 19:07 jiwoongim