Jason Senthil

Fix/syncmode

2

Fixes #4008 The goal is to change the boolean flags FastSync, FastBlocks, SnapSync to an enumeration within the SyncConfig, and mark those boolean flags as obsolete. An enumeration was added,...

remove size_bounds arg from Timer

1

Summary: happened across the Timer class which had documentation for arg which it does not have Reviewed By: galrotem Differential Revision: D55167092

update checkpoint docs fsdp state dict augment

1

Differential Revision: D54640060

forward process group properly to rank 0 decorator

2

Differential Revision: D54544617

update auto unit docs on precision

1

Summary: clarifies the precision docstring in AutoUnit Reviewed By: galrotem Differential Revision: D54344882

move checkpointing FAQ to new debugging section

2

Summary: since the questions were related to debugging anyway Reviewed By: galrotem Differential Revision: D54208613

bump TorchTNT to version 0.2.3

1

Summary: bump to new version Differential Revision: D54137873

add file arg to tqdm utils / callbacks

4

Summary: # Context tqdm by default prints to stderr. This can be unintuitive location for the progress bar to appear # This Diff Adds `file` arg so users can pass...

redo torchvision dependency fix

1

Summary: previous diff D53456248 mistakenly pinned the torchvision version in the mnist example. We want to pin the version in the dev-requirements instead to fix the following error: ``` ====================================...

remove progress from replicated in TSS if ["**"] passed in

1

Differential Revision: D53284173