临终遗言功能相比故障恢复功能区别在于发生故障时会保存断点处的临终ckpt文件,重训练时直接加载临终ckpt文件,实现断点续训。
cd /data/atlas_dls/public/code/ResNet50_for_MindSpore_2.0.0-alpha_code_endpoint/output/checkpoint/ckpt_0
ll ./
total 800448 drwx------ 2 root root 4096 Mar 22 19:34 ./ drwx------ 18 root root 4096 Mar 22 19:21 ../ -r-------- 1 root root 204715348 Mar 22 19:24 resnet-10_24.ckpt -r-------- 1 root root 204715348 Mar 22 19:34 resnet_1-12_24_breakpoint.ckpt -r-------- 1 root root 204715348 Mar 22 19:34 resnet-12_24_breakpoint.ckpt -r-------- 1 root root 204715348 Mar 22 19:24 resnet-5_24.ckpt -rw------- 1 root root 767018 Mar 22 19:23 resnet-graph.meta