0

mpi_train.sh 有以下行:

mpiexec.hydra -prepend-rank -host node11 -n 16

在并行 caffe 中运行 examples/cifar10/mpi_train_quick.sh 时,出现以下错误:

F0521 18:33:32.101737 16179 caffe.cpp:293] This MPI version is NOT support multi-thread!
*** Check failure stack trace: ***
@ 0x7f8367e83daa (unknown)
@ 0x7f8367e83ce4 (unknown)
@ 0x7f8367e836e6 (unknown)
@ 0x7f8367e86687 (unknown)
@ 0x410ac8 main
@ 0x7f8364b76ec5 (unknown)
@ 0x414e77 (unknown)
@ (nil) (unknown)
Aborted (core dumped)

我已经安装了 mpich-3.1.4 版本。当我执行 which mpirun, which mpiexec, which mpicc 时,我从所有这些中获取 mpich-3.1.4 文件夹的路径此外,系统同时安装了 Open-Mpi 和 MPICH-3.1.4,因为 open-mpi 自动安装在安装 opencv 时,这是并行 caffe 的先决条件。如何解决此错误?

4

0 回答 0