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

Jump to bottom

Add option to disable distributed parameters in distributed Adam optimizer #5685

Merged
merged 6 commits into from
Jan 31, 2023
Merged

Add option to disable distributed parameters in distributed Adam optimizer #5685

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
76bb39e
Add option to run dist Adam without distributed params
timmoon10 Dec 21, 2022
affedef
Merge branch 'main' into dist-adam-without-dist-params
timmoon10 Dec 23, 2022
b7bc08e
Merge branch 'main' into dist-adam-without-dist-params
timmoon10 Jan 5, 2023
b72f44a
Merge branch 'main' into dist-adam-without-dist-params
timmoon10 Jan 20, 2023
e857bde
Fix bug in grad clipping when dist Adam has redundant params
timmoon10 Jan 21, 2023
7f3a0a2
Merge branch 'main' into dist-adam-without-dist-params
okuchaiev Jan 31, 2023
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: 2 additions & 0 deletions nemo/collections/nlp/modules/common/megatron/clip_grads.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -190,6 +190,8 @@ def clip_grad_norm_distributed_optimizer(optimizer, max_norm, norm_type=2):
# Compute grad norm
# Note: Compute norm of local grads and sum over all procs
grad_norm_sq = optimizer._local_grad_norm(parameters=params_for_norm, norm_type=norm_type)
if optimizer.redundant_size > 1:
grad_norm_sq /= optimizer.redundant_size
Comment on lines +193 to +194
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do all optimizers have this attribute? Can we do a safer check here?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This grad clipping function guarantees that it's just dealing with the distributed optimizer:

NeMo/nemo/collections/nlp/modules/common/megatron/clip_grads.py

Line 177 in e626026

assert isinstance(optimizer, DistributedFusedAdam)

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've also been thinking about moving this logic to the distributed optimizer wrapper at:

NeMo/nemo/core/optim/distributed_adam.py

Line 21 in fd6bb1d

class MegatronDistributedFusedAdam(DistributedFusedAdam):

I've been working on supporting BF16 grad reductions, so perhaps this work can be folded into that PR.

torch.distributed.all_reduce(
grad_norm_sq, op=torch.distributed.ReduceOp.SUM,
)
Expand Down
9 changes: 8 additions & 1 deletion nemo/core/optim/distributed_adam.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,16 +12,23 @@
# See the License for the specific language governing permissions and
# limitations under the License.

import torch
from apex.contrib.optimizers.distributed_fused_adam import DistributedFusedAdam
from apex.transformer import parallel_state


# Wrapper class that supports main_grad buffer
# Note: main_grad buffer is used for O2-style optimizations
class MegatronDistributedFusedAdam(DistributedFusedAdam):
def __init__(self, *args, **kwargs):
def __init__(self, *args, disable_distributed_parameters=False, **kwargs):
if 'process_group' not in kwargs and not parallel_state.is_unitialized():
kwargs['process_group'] = parallel_state.get_data_parallel_group()
if disable_distributed_parameters:
world_size = torch.distributed.get_world_size()
rank = torch.distributed.get_rank()
self_groups = [torch.distributed.new_group(ranks=[i]) for i in range(world_size)]
kwargs['distributed_process_group'] = self_groups[rank]
kwargs['redundant_process_group'] = kwargs['process_group']
super().__init__(*args, **kwargs)

def _make_post_backward_hook(self, param, param_group_id, param_id):
Expand Down