tensorflow+MPI多进程报错的解决办法:Signal: Segmentation fault (11)Signal code: Address not mapped (1)

在使用TensorFlow和MPI进行多进程运算时遇到Segmentation fault (11)错误。问题源于自定义梯度函数在MPI环境中导致的问题,而非函数本身。通过检查发现OpenMPI版本较低,升级到最新版(4.1.4)解决了问题,无需sudo权限,只需通过conda安装和重装。

摘要生成于 C知道 ,由 DeepSeek-R1 满血版支持, 前往体验 >

tensorflow+MPI跑一个任务。遇到如下错误:

[cdc-24:02776] *** Process received signal ***
[cdc-24:02776] Signal: Segmentation fault (11)
[cdc-24:02776] Signal code: Address not mapped (1)
[cdc-24:02776] Failing at address: 0x55ea3fd8c7f0
[cdc-24:02771] *** Process received signal ***
[cdc-24:02771] Signal: Segmentation fault (11)
[cdc-24:02771] Signal code: Address not mapped (1)
[cdc-24:02771] Failing at address: 0x1d0
[cdc-24:02778] *** Process received signal ***
[cdc-24:02778] Signal: Segmentation fault (11)
[cdc-24:02778] Signal code: Address not mapped (1)
[cdc-24:02778] Failing at address: 0x5569a83df7f0
[cdc-24:02773] *** Process received signal ***
[cdc-24:02776] [ 0] /lib/x86_64-linux-gnu/libpthread.so.0(+0x11390)[0x7f3fa8e7b390]
[cdc-24:02776] [ 1] [cdc-24:02778] [ 0] /lib/x86_64-linux-gnu/libpthread.so.0(+0x11390)[0x7fd44ecf9390]
[cdc-24:02778] [ 1] /usr/lib/openmpi/lib/openmpi/mca_btl_vader.so(mca_btl_vader_poll_handle_frag+0x154)[0x7fd4185c18d4]
评论 1
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值