Skip to content

fix loading dcp OOM #1747

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
zjjott wants to merge 1 commit into
base: main
Choose a base branch
from
+3 −12
Draft

fix loading dcp OOM #1747

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion megatron/core/optimizer/distrib_optimizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -771,7 +771,7 @@ def make_needed_groups(param_group):
# Allocate dummy tensors.
numel = len(param_range_map["gbuf_world"])
init_shard = lambda dtype=torch.float32: torch.empty(
(numel,), dtype=dtype, device=torch.cuda.current_device()
(numel,), dtype=dtype, device="cpu"
)

# For precision_aware_optimizer, the empty tensors should also be
Expand Down
13 changes: 2 additions & 11 deletions megatron/core/transformer/mlp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -340,17 +340,8 @@ def sh_ten_build_fn(

def sh_ten_merge_fn(sub_state_dict):
with torch.no_grad():
try:
return torch.cat(sub_state_dict)
except (RuntimeError, torch.cuda.OutOfMemoryError) as e:
logger.warning(
f"CUDA OutOfMemoryError encountered during tensors merging."
f" Switching to CPU merge. (Error: {e})"
)
merged_sub_state_dict = torch.cat([t.cpu() for t in sub_state_dict])
gc.collect()
torch.cuda.empty_cache()
return merged_sub_state_dict
merged_sub_state_dict = torch.cat([t.cpu() for t in sub_state_dict])
return merged_sub_state_dict

return ShardedTensorFactory(
original_sh_ten.key,
Expand Down