pytorch / ignite

High-level library to help with training and evaluating neural networks in PyTorch flexibly and transparently.
https://pytorch-ignite.ai
BSD 3-Clause "New" or "Revised" License
4.51k stars 612 forks source link

Extension of `ignite.distributed.utils.broadcast` for pathlib.Path datatype #3025

Open guptaaryan16 opened 1 year ago

guptaaryan16 commented 1 year ago

🚀 Feature

I would like to request for extension of ignite.distributed.utils.broadcast for Path datatype as it is frequently used for distributed training and can be very useful for designing DL pipeline based projects. The API should work like this

from pathlib import Path
import ignite.distributed as idist

path = Path("./log")

idist.broadcast(path, src=0)

Further if possible we can also provide support for more data types as suggested by others. Also if possible I would like to take up this issue. @vfdev-5 @sadra-barikbin

sadra-barikbin commented 1 year ago

@guptaaryan16 , thank you for the suggestion. Couldn't we broadcast the path as a string which is supported now ?

guptaaryan16 commented 1 year ago

@sadra-barikbin Yeah it does but I was hoping more data types can be supported as it seems possible in the codebase