fix(zero): warn when process group silently falls back to single-rank#8117
Open
NovusEdge wants to merge 1 commit into
Open
fix(zero): warn when process group silently falls back to single-rank#8117NovusEdge wants to merge 1 commit into
NovusEdge wants to merge 1 commit into
Conversation
When WORLD_SIZE env var indicates multi-GPU but the process group has world_size=1, emit a warning explaining the likely cause and fix. This catches the silent failure where zero.Init sees world_size=1 because torch.distributed wasn't initialized before model construction, leading to each rank loading the full model and OOM. Closes deepspeedai#8084 Signed-off-by: NovusEdge <novusedge0@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds a warning when
zero.Initdetects thatWORLD_SIZEenv var suggests multi-GPU but the actual process group hasworld_size=1.This catches the sneaky case where torch.distributed wasnt initialized before model construction, so each rank loads the full model instead of sharding -> boom, OOM :D
The problem
When using
deepspeed --num_gpus 8but callingfrom_pretrained()beforedeepspeed.init_distributed(), zero.Init silently falls back to single-rank mode. No warning, just an OOM that looks like "model too big" when it actually fits fine if sharded properly.The fix
Check if
WORLD_SIZE> 1 butdp_world_size== 1 after process group setup, and warn loudly with a clear explanation of whats happening and how to fix it.Closes #8084