Maskrcnn-benchmark: Pycharm configuration - multi GPU training

[barakhi]

❓ Questions and Help

Hi,
I'm trying to run multi-GPU training in Pycharm and can't figure out how to configure my run.
I've tried:
Module name (instead of script path): torch.distributed.launch
Parameters: --nproc_per_node=2 tools/train_net.py --config-file ...
with sum variations and with no success.

How should I configure my run ?
Thanks

[miguelvr]

Here you go

[fmassa]

Closing thanks to @miguelvr suggestion

[barakhi]

Hi,
I tried Miguel's suggestion - but got:
"No module named tools/train_net.py"

Have I missed anything?

Thanks,

[miguelvr]

tools/train.net is not a module so you are not filling the correct fields

Look at the picture

[barakhi]

This is my config:

Where is my mistake? Can't find it

[miguelvr]

that indeed looks ok... can you show me the output of the terminal?

[barakhi]

It is working well from the terminal.
When I'm trying to run the configuration - It is also working.
The problem only occurs while debugging the configuration.
I don't understand the reason, but I guess it's a question to PyCharm - not to you.
Thanks, anyway

[miguelvr]

you are better off debugging with a single GPU, because torch.launch.distributed is spawning multiple processes. Not sure if pycharm can handle that