Skip to content
New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

#

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Jump to bottom

Changes in create_optimizer to support tensor parallelism with SMP #16880

Merged
merged 2 commits into from
Apr 22, 2022
Merged

Changes in create_optimizer to support tensor parallelism with SMP #16880

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1626854
changes in create optimizer to support tensor parallelism with SMP
Apr 20, 2022
05cb2d5
Update src/transformers/trainer.py
cavdard Apr 22, 2022
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
10 changes: 6 additions & 4 deletions src/transformers/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -843,16 +843,18 @@ def create_optimizer(self):
We provide a reasonable default that works well. If you want to use something else, you can pass a tuple in the
Trainer's init through `optimizers`, or subclass and override this method in a subclass.
"""
opt_model = self.model_wrapped if is_sagemaker_mp_enabled() else self.model

if self.optimizer is None:
decay_parameters = get_parameter_names(self.model, [nn.LayerNorm])
decay_parameters = get_parameter_names(opt_model, [nn.LayerNorm])
decay_parameters = [name for name in decay_parameters if "bias" not in name]
optimizer_grouped_parameters = [
{
"params": [p for n, p in self.model.named_parameters() if n in decay_parameters],
"params": [p for n, p in opt_model.named_parameters() if n in decay_parameters],
"weight_decay": self.args.weight_decay,
},
{
"params": [p for n, p in self.model.named_parameters() if n not in decay_parameters],
"params": [p for n, p in opt_model.named_parameters() if n not in decay_parameters],
"weight_decay": 0.0,
},
]
Expand All @@ -872,7 +874,7 @@ def create_optimizer(self):

manager = bitsandbytes.optim.GlobalOptimManager.get_instance()

for module in self.model.modules():
for module in opt_model.modules():
if isinstance(module, nn.Embedding):
manager.register_module_override(module, "weight", {"optim_bits": 32})
logger.debug(f"bitsandbytes: will optimize {module} in fp32")
Expand Down