Skip to content

M
models

PaddlePaddle / models
大约 2 年 前同步成功

通知 232
Star 6828
Fork 2962
M
models

Further speedup DS2 training by a) turning on FLAG_rnn_use_batch 2) improve data pipeline !359

!359 已合并 由 saxon_zh@saxon_zh 创建

Created by: xinghai-sun

Resolved #358 (closed)

  1. Setting FLAG_rnn_use_batch to True (default is False) might got more than 2X speedup for paddle.layer.recurrent layer.
  2. Improve data pipeline: add a separate thread to read data from Manager.Queue and re-save to faster Queue.Queue in xmap_reader_mp. This save up 10% ~ 15% training time.