Skip to content

Use odd shape tensor to represent parameter data in partitioned state #981

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

Merged
cli99 merged 4 commits into from
Apr 21, 2021
Merged

Use odd shape tensor to represent parameter data in partitioned state #981

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
250d0a2
use wierd shaped tensor to avoid silent failures when not registering…
cli99 Apr 20, 2021
060285b
fix typo
cli99 Apr 20, 2021
2e5a950
Merge branch 'master' into cheng/partitioned-param-representation
tjruwase Apr 20, 2021
174dba3
Merge branch 'master' into cheng/partitioned-param-representation
cli99 Apr 21, 2021
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
6 changes: 4 additions & 2 deletions deepspeed/runtime/zero/partition_parameters.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,7 @@
from ..config import DeepSpeedConfig

param_count = 0
partitioned_param_data_shape = [1]


def print_rank_0(message, debug=False, force=False):
Expand Down Expand Up @@ -631,7 +632,8 @@ def _partition_param(self, param, buffer=None, has_been_updated=False):
f'Before partitioning param {param.ds_id} {param.shape}',
force=False)
#param.data does not store anything meaningful in partitioned state
param.data = torch.ones(1).half().to(param.device)
param.data = torch.ones(partitioned_param_data_shape).half().to(
param.device)
see_memory_usage(f'After partitioning param {param.ds_id} {param.shape}',
force=False)

Expand Down Expand Up @@ -712,7 +714,7 @@ def _partition_param(self, param, buffer=None, has_been_updated=False):

see_memory_usage(f'Before partitioning param {param.ds_id} {param.shape}',
force=False)
param.data = torch.ones(1).half().to(param.device)
param.data = torch.ones(partitioned_param_data_shape).half().to(param.device)
see_memory_usage(f'After partitioning param {param.ds_id} {param.shape}',
force=False)

Expand Down
4 changes: 2 additions & 2 deletions tests/unit/test_zero_context.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@
import pytest

import deepspeed
from deepspeed.runtime.zero.partition_parameters import ZeroParamStatus
from deepspeed.runtime.zero.partition_parameters import ZeroParamStatus, partitioned_param_data_shape

from common import distributed_test

Expand All @@ -32,7 +32,7 @@ def test_scatter_gather():
with deepspeed.zero.Init():
l = torch.nn.Linear(6, 3)
assert l.weight.ds_status == ZeroParamStatus.NOT_AVAILABLE
assert l.weight.numel() == 1
assert l.weight.shape == torch.Size(partitioned_param_data_shape)

# Ensure there is no impact outside the context
l2 = torch.nn.Linear(6, 3)
Expand Down