🐛 Describe the bug

When calling dist.initialize_dist, I can specify a dist_timeout argument.

When training with FSDP and device_mesh, I want to call from torch.distributed._tensor import init_device_mesh and pass the device_mesh into FSDP. However, it seems that the process groups created do not respect dist_timeout through this process

Versions

Torch nightly 1/10/2024

cc @mrshenli @pritamdamania87 @zhaojuanmao @satgera @rohan-varma @gqchen @aazzolini @osalpekar @jiayisuse @H-Huang @kwen2501 @awgu @penguinwu @fegin @XilunWu @wanchaol @fduwjj @wz337 @tianyu-l @wconstab @yf225 @LucasLLC