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

fixes in _partition_param_sec function #5613

Merged
merged 1 commit into from
Jun 11, 2024
Merged

fixes in _partition_param_sec function #5613

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
5 changes: 4 additions & 1 deletion deepspeed/runtime/zero/partition_parameters.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1664,6 +1664,8 @@ def _partition_param_sec(self, param, buffer=None, has_been_updated=False):
##support for NVME secondary param offload
#print_rank_0(f"SEC Param id {param.ds_id} status is {param.ds_status}", force=True)
if param.ds_status is ZeroParamStatus.AVAILABLE:
if param.ds_secondary_tensor is not None and not has_been_updated: ##param already partitioned
return
#check padding
tensor_size = self._aligned_size(param)
partition_size = tensor_size // self.dp_world_size
Expand Down Expand Up @@ -1702,7 +1704,8 @@ def _partition_param_sec(self, param, buffer=None, has_been_updated=False):
sec_numel).copy_(one_dim_param.narrow(0, secondary_start, sec_numel))

# TODO: This is a temporary fix to avoid the issue that 2nd tensor all-gather happens before 2nd tensor partition is done
get_accelerator().current_stream().synchronize()
if not get_accelerator().resolves_data_dependency():
get_accelerator().current_stream().synchronize()

print_rank_0(f"{param.ds_id} partitioned type {param.dtype} dev {param.device} shape {param.shape}",
force=False)
Expand Down
Loading