how i use docker mode or seven mode to train using muti-GPUs about pocketflow HOT 5 CLOSED

@jiayiliu
"Seven" is only available within Tencent. For users outsides Tencent, please use the local & docker mode.

from pocketflow.

1. The error message seems related to the docker configuration, instead of PocketFlow.
2. To prepare the training data for the docker mode, just make sure the data_dir_local_ilsvrc12 entry in the path.conf file is correctly configured to match your ILSVRC-12 data set. The execution script, scripts/run_docker.sh, will mount the local data directory to /opt/ml/data in the docker environment. You do not need to modify the data_dir_docker_ilsvrc12 entry.

from pocketflow.

In scripts/run_docker.sh, NV_GPU=xxx docker.oa.com/g_tfplus/horovod:python3.5 bash. Should I use "uber/horovod" instead of "docker.oa.com/g_tfplus/horovod:python3.5", because i have pulled image uber/horovod .

from pocketflow.

Yes, you should use "uber/horovod" instead of "docker.oa.com/g_tfplus/horovod:python3.5".

from pocketflow.

Any reference for the distributed learning platform "Seven"? I didn't find it somehow... Thanks

from pocketflow.