search

nouhadziri / THRED

The implementation of the paper "Augmenting Neural Response Generation with Context-Aware Topical Attention"
https://arxiv.org/abs/1811.01063
MIT License
111 stars 25 forks source link

GPU-Util is low #9

Closed shark-3 closed 4 years ago

shark-3 commented 5 years ago

The GPU-Util is very low ( mostly 0 ) and the CPU utilizes above 1000%. I installed tensorflow-gpu==1.8. And I used CUDA_VISIBLE_DEVICES to assign GPU to training.

ehsk commented 5 years ago

Seems odd! Which model are you running? Also, can you make sure that you don't have tensorflow-cpu installed in your environment?

shark-3 commented 5 years ago

Seems odd! Which model are you running? Also, can you make sure that you don't have tensorflow-cpu installed in your environment?

I used baseline model HRED and I am sure that I installed tensorflow-gpu=1.8.0. Sometimes the GPU-Util will be above 20%, but 0% most of the time. UQ{)8YFY_BFGYCZY3`9X`DN

ehsk commented 5 years ago

Could you turn on "log_device" in the config file and share the log with us?

shark-3 commented 5 years ago

Could you turn on "log_device" in the config file and share the log with us?

I sent it to your email. Thank you very much for your reply!