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

fix cpm 1x1 for FP32 #215

Merged
merged 1 commit into from
Aug 29, 2023
Merged

fix cpm 1x1 for FP32 #215

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 3 additions & 1 deletion training/benchmarks/cpm/pytorch/train/evaluator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,9 @@ def evaluate(self, trainer):
all_losses.append(loss.item())

preds = torch.argmax(output, -1)
if isinstance(model.module, FP16_Module):
if not hasattr(model, "module"):
embeddings = model.word_embeddings.weight
elif isinstance(model.module, FP16_Module):
embeddings = model.module.module.word_embeddings.weight
else:
embeddings = model.module.word_embeddings.weight
Expand Down
5 changes: 4 additions & 1 deletion training/benchmarks/cpm/pytorch/train/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -142,7 +142,10 @@ def train_one_step(self, batch, no_model_batch):

# calculate output
preds = torch.argmax(output, -1)
if isinstance(self.model.module, FP16_Module):

if not hasattr(self.model, "module"):
embeddings = self.model.word_embeddings.weight
elif isinstance(self.model.module, FP16_Module):
embeddings = self.model.module.module.word_embeddings.weight
else:
embeddings = self.model.module.word_embeddings.weight
Expand Down
19 changes: 10 additions & 9 deletions training/benchmarks/cpm/pytorch/train/trainer_adapter.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,16 +38,17 @@ def create_optimizer(config, model: nn.Module) -> Optimizer:

def model_to_fp16(config, model: nn.Module,
optimizer: Optimizer) -> Tuple[nn.Module, Optimizer]:
model = FP16_Module(model)
args = config
optimizer = FP16_Optimizer(optimizer,
static_loss_scale=args.loss_scale,
dynamic_loss_scale=args.dynamic_loss_scale,
dynamic_loss_args={
'scale_window': args.loss_scale_window,
'min_scale': args.min_scale,
'delayed_shift': args.hysteresis
})
if args.fp16:
model = FP16_Module(model)
optimizer = FP16_Optimizer(optimizer,
static_loss_scale=args.loss_scale,
dynamic_loss_scale=args.dynamic_loss_scale,
dynamic_loss_args={
'scale_window': args.loss_scale_window,
'min_scale': args.min_scale,
'delayed_shift': args.hysteresis
})

return model, optimizer

Expand Down