No such file or directory

[YuNaruto]

when i run this: python -m runx.runx scripts/dump_folder.yml -i , go wrong . but the path is available. how this?

[YuNaruto]

@ajtao the path logs/dump_folder/hidden-puffin_2021.04.27_17.17/code is available

[atharvas]

Notice that we're cd-ing into the logs/ directory but the command executed, as defined in scripts/dump_folder.yml is:

python -m torch.distributed.launch --nproc_per_node=1 train.py

However, train.py is defined at the project directory level. Changing the relative path to an absolute path should fix this issue. That is:

CMD: "python -m torch.distributed.launch --nproc_per_node=1 {absolute path to train.py}"

In my case this was:

CMD: "python -m torch.distributed.launch --nproc_per_node=1 /home/ubuntu/semantic-segmentation/train.py"

[ajtao]

There's something wrong with (both) of your setups. What runx is attempting to do is:

1. Copy the codebase into a new 'run' directory under LOGROOT (defined in .runx). What @NarutoZhao's output showed is that this copy is failing. You should try to diagnose that problem. Try to define LOGROOT with an absolute path.
2. @atharvas Your comment is wrong. train.py is invoked after runx cd's into the run directory, where the code has just been copied. So you should not need to refer to the absolute path of train.py. In fact, the purpose of copying the code to the run directory is to serve as both (a) an archive of the state of the code and (b) to allow you to further modify your project repo after launching jobs.

[msseibel]

See also #86