Skip to content
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

add check_meta and sync_data for to_global #530

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

add check_meta and sync_data for to_global #530

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
9 changes: 5 additions & 4 deletions libai/data/structures.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@ class DistTensorData:
placement_idx: int = 0

# Tensor-like methods
def to_global(self, sbp=None, placement=None, device_type="cuda"):
def to_global(self, sbp=None, placement=None, device_type="cuda", check_meta=True, sync_data=True):
if sbp is not None:
self.sbp = sbp
else:
Expand All @@ -47,7 +47,7 @@ def to_global(self, sbp=None, placement=None, device_type="cuda"):
self.sbp = dist.get_nd_sbp(sbp_list)

if placement is not None:
self.tensor = self.tensor.to_global(sbp=self.sbp, placement=placement)
self.tensor = self.tensor.to_global(sbp=self.sbp, placement=placement, check_meta=check_meta, sync_data=sync_data)
else:
# Convert local tensor to global tensor with default setting,
# if the placement parameter is not provided.
Expand All @@ -62,10 +62,11 @@ def to_global(self, sbp=None, placement=None, device_type="cuda"):
# by the fist device group, in case that each device group containg
# some random augmentations to the tensors without setting the same global seed.
main_placement = dist.get_layer_placement(0, device_type)
self.tensor = self.tensor.to_global(sbp=self.sbp, placement=main_placement)
self.tensor = self.tensor.to_global(sbp=self.sbp, placement=main_placement, check_meta=check_meta, sync_data=sync_data)
if self.placement_idx != 0:
self.tensor = self.tensor.to_global(
placement=dist.get_layer_placement(self.placement_idx, device_type)
placement=dist.get_layer_placement(self.placement_idx, device_type),
check_meta=check_meta, sync_data=sync_data
)

@staticmethod
Expand Down
4 changes: 3 additions & 1 deletion libai/engine/default.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -496,6 +496,8 @@ def get_batch(
data: Instance,
input_placement_device: str = "cuda",
mixup_func: Optional[Callable] = None,
check_meta: bool=True,
sync_data: bool=True,
):
"""
Convert batched local tensor to distributed tensor for model step running.
Expand All @@ -516,7 +518,7 @@ def get_batch(

ret_dict = {}
for key, value in data.get_fields().items():
value.to_global(device_type=input_placement_device)
value.to_global(device_type=input_placement_device, check_meta=check_meta, sync_data=sync_data)
ret_dict[key] = value.tensor
return ret_dict

Expand Down