Skip to content

set adamw_mode default true (follows FusedAdam and < 0.3.11 logic) #844

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
jeffra merged 4 commits into from
Mar 11, 2021
Merged

set adamw_mode default true (follows FusedAdam and < 0.3.11 logic) #844

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
37e4e74
set adamw_mode default true (follows FusedAdam and < 0.3.11 logic)
jeffra Mar 9, 2021
c4df508
refactor of adam(w) logic, still same effect as before but hopefully …
jeffra Mar 9, 2021
a23891a
add unit tests sweeping the space of adam/adamw
jeffra Mar 11, 2021
2e391d2
Merge branch 'master' into jeffra/adamw_default_true
jeffra Mar 11, 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
4 changes: 4 additions & 0 deletions deepspeed/runtime/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,10 @@
# extra optimizer parameters for adam/adamw
TORCH_ADAM_PARAM = "torch_adam"

# default to adamw logic for adam/adamw optimizers unless user explictly opts out
ADAM_W_MODE = "adam_w_mode"
ADAM_W_MODE_DEFAULT = True


class DeepSpeedConfigError(Exception):
pass
Expand Down
8 changes: 4 additions & 4 deletions deepspeed/runtime/engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@
from deepspeed.runtime.fp16.unfused_optimizer import FP16_UnfusedOptimizer
from deepspeed.runtime.config import DeepSpeedConfig, DEEPSPEED_OPTIMIZERS, \
ADAM_OPTIMIZER, ADAMW_OPTIMIZER, LAMB_OPTIMIZER, ONEBIT_ADAM_OPTIMIZER, \
TORCH_ADAM_PARAM
TORCH_ADAM_PARAM, ADAM_W_MODE, ADAM_W_MODE_DEFAULT

from deepspeed.runtime.dataloader import DeepSpeedDataLoader
from deepspeed.runtime.constants import \
Expand Down Expand Up @@ -640,10 +640,10 @@ def _configure_basic_optimizer(self, model_parameters):

if self.optimizer_name() in [ADAM_OPTIMIZER, ADAMW_OPTIMIZER]:
torch_adam = optimizer_parameters.pop(TORCH_ADAM_PARAM, False)
adam_w_mode = self.optimizer_name() == ADAMW_OPTIMIZER
adam_w_mode = optimizer_parameters.get(ADAM_W_MODE, ADAM_W_MODE_DEFAULT)
Comment thread
jeffra marked this conversation as resolved.
Outdated
# zero-offload torch-adam adam_w_mode optimizer
# T|F T T torch.optim.AdamW
# T|F T F torch.optim.Adam
# T|F T T torch.optim.Adam
# T F T|F DeepSpeedCPUAdam(adam_w_mode)
# F F T|F FusedAdam(adam_w_mode)
if torch_adam:
Expand All @@ -658,7 +658,7 @@ def _configure_basic_optimizer(self, model_parameters):
**optimizer_parameters,
adamw_mode=adam_w_mode)
else:
optimizer_parameters['adam_w_mode'] = adam_w_mode
optimizer_parameters[ADAM_W_MODE] = adam_w_mode
optimizer = FusedAdam(model_parameters, **optimizer_parameters)

elif self.optimizer_name() == LAMB_OPTIMIZER:
Expand Down