multi-node distributed training

[gogogwwb]

Is there an example of multi-node distributed training?

[BoxiangW]

We have provided ColossalAI-Examples for various examples regarding our features and training processes for different tasks.

[gogogwwb]

other machine seems to be running in the background.

[gogogwwb]

Is this a parameter setting incorrect?

[FrankLeeeee]

When multi-machine distributed training, only one machine is running, why is this?Need to modify the code?resnet，pipeline_parallel

To do a simple check, you can write a simple test file like

pritn('launched')

then run `colossalai run --nproc_per_node 2 --host host1,host2 --master_addr host1 test.py.

You should be able to see 4 lines of launched.

[FrankLeeeee]

Is this a parameter setting incorrect?

For this one, seems that your gloo has some issue. You can check your network interface by ifconfig and look for the one for your ethernet. Lastly, run GLOO_SOCKET_IFNAME=<interface-name> colossalai run ...

[gogogwwb]

Is the reason for this error because of the config file?