Skip to content

EMA: fix state_dict() and load_state_dict() & add cur_decay_value #2146

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
merged 5 commits into from
Feb 8, 2023
Merged

EMA: fix state_dict() and load_state_dict() & add cur_decay_value #2146

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
c3e70c3
EMA: fix `state_dict()` & add `cur_decay_value`
chenguolin Jan 28, 2023
e60364c
EMA: fix a bug in `load_state_dict()`
chenguolin Jan 29, 2023
83c5093
del train_unconditional_ort.py
chenguolin Feb 7, 2023
81bdae0
Merge branch 'main' of https://github.com/chenguolin/diffusers into main
chenguolin Feb 7, 2023
14d6b79
Merge branch 'huggingface:main' into main
chenguolin Feb 7, 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: 1 addition & 1 deletion examples/unconditional_image_generation/train_unconditional.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -526,7 +526,7 @@ def transform_images(examples):

logs = {"loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0], "step": global_step}
if args.use_ema:
logs["ema_decay"] = ema_model.decay
logs["ema_decay"] = ema_model.cur_decay_value
progress_bar.set_postfix(**logs)
accelerator.log(logs, step=global_step)
progress_bar.close()
Expand Down
4 changes: 3 additions & 1 deletion src/diffusers/training_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -122,6 +122,7 @@ def __init__(
self.inv_gamma = inv_gamma
self.power = power
self.optimization_step = 0
self.cur_decay_value = None # set in `step()`

def get_decay(self, optimization_step: int) -> float:
"""
Expand Down Expand Up @@ -163,6 +164,7 @@ def step(self, parameters: Iterable[torch.nn.Parameter]):

# Compute the decay factor for the exponential moving average.
decay = self.get_decay(self.optimization_step)
self.cur_decay_value = decay
one_minus_decay = 1 - decay

for s_param, param in zip(self.shadow_params, parameters):
Expand Down Expand Up @@ -208,7 +210,7 @@ def state_dict(self) -> dict:
# https://pytorch.org/tutorials/beginner/saving_loading_models.html#what-is-a-state-dict
return {
"decay": self.decay,
"min_decay": self.decay,
"min_decay": self.min_decay,
"optimization_step": self.optimization_step,
"update_after_step": self.update_after_step,
"use_ema_warmup": self.use_ema_warmup,
Expand Down