Skip to content

disable kernel promotion for amp traning #1085

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

Open
wants to merge 2 commits into
base: develop
Choose a base branch
from
Open

disable kernel promotion for amp traning #1085

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4b1f44b
disable kernel promotion for amp traning
zhangting2020 May 15, 2023
8bc6310
Merge branch 'develop' into develop
zhangting2020 May 16, 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
13 changes: 9 additions & 4 deletions ppfleetx/core/engine/eager_engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -142,6 +142,7 @@ def configure_optimizers(self):

self._amp_dtype = amp_config.get('dtype', 'float16')
self._amp_level = amp_config.get('level', 'O2')
self._amp_use_promote = amp_config.get('use_promote', False)
self._use_main_grad = amp_config.get('use_main_grad', False)
self._scale_loss = amp_config['scale_loss']
self._custom_black_list = amp_config['custom_black_list']
Expand Down Expand Up @@ -510,7 +511,8 @@ def _fit_impl(self, batch):
custom_black_list=self._custom_black_list,
custom_white_list=self._custom_white_list,
dtype=self._amp_dtype,
level=self._amp_level):
level=self._amp_level,
use_promote=self._amp_use_promote):
batch = self._module.model._prepare_training(
batch, self._optimizer, self._lr_scheduler)
loss = self._module.model.forward_backward_pipeline(
Expand All @@ -537,7 +539,8 @@ def _model_forward_backward(self, batch):
custom_black_list=self._custom_black_list,
custom_white_list=self._custom_white_list,
dtype=self._amp_dtype,
level=self._amp_level):
level=self._amp_level,
use_promote=self._amp_use_promote):
loss = self._module.training_step(micro_batch)

if self._amp_enable and self._amp_dtype == "float16":
Expand Down Expand Up @@ -648,7 +651,8 @@ def _evaluate_impl(self, batch):
custom_black_list=self._custom_black_list,
custom_white_list=self._custom_white_list,
dtype=self._amp_dtype,
level=self._amp_level):
level=self._amp_level,
use_promote=self._amp_use_promote):
if self._pp_degree == 1:
loss = self._module.validation_step(batch)
else:
Expand Down Expand Up @@ -706,7 +710,8 @@ def _predict_impl(self, batch):
custom_black_list=self._custom_black_list,
custom_white_list=self._custom_white_list,
dtype=self._amp_dtype,
level=self._amp_level):
level=self._amp_level,
use_promote=self._amp_use_promote):
if self._pp_degree == 1:
loss = self._module.test_step(batch)
else:
Expand Down