Skip to content

Latest commit

 

History

History
18 lines (14 loc) · 745 Bytes

README.md

File metadata and controls

18 lines (14 loc) · 745 Bytes

pytorch-rnn-benchmark


A program to test performance of several RNNs(realized by Pytorch) on different hardware platforms, providing benchmark for further optimization.

Requirements


  • Pytorch

Usage


bash run.sh platform-name [train] [daily]

platform-name can be 'bdw', 'skx', 'knm', 'knl', 'gpu'.
If you want to test performance of training, please add parameter train.
If you want to run daily test(just test 2 size), please add parameter daily.

Benchmark size info

For LSTM/RNN/GRU, we will follow the size of Baidu DeepBench. And for LSTM we wiil add OpenNMT size as an supplement.