aljaz/DALLE2-pytorch
1
1
Fork 0
mirror of https://github.com/lucidrains/DALLE2-pytorch.git synced 2025-12-19 09:44:19 +01:00
Code Issues Packages Projects Releases Wiki Activity

foolproof sampling for decoder to always use eval mode (and restore training state afterwards)

Browse Source
This commit is contained in:
Phil Wang
2022-07-13 10:21:00 -07:00
parent cc0f7a935c
commit b2073219f0
2 changed files with 9 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
dalle2_pytorch/trainer.py
View File

@@ -673,8 +673,14 @@ class DecoderTrainer(nn.Module):
def sample(self, *args, **kwargs): def sample(self, *args, **kwargs):
distributed = self.accelerator.num_processes > 1 distributed = self.accelerator.num_processes > 1
base_decoder = self.accelerator.unwrap_model(self.decoder) base_decoder = self.accelerator.unwrap_model(self.decoder)
was_training = base_decoder.training
base_decoder.eval()
if kwargs.pop('use_non_ema', False) or not self.use_ema: if kwargs.pop('use_non_ema', False) or not self.use_ema:
return base_decoder.sample(*args, **kwargs, distributed = distributed) out = base_decoder.sample(*args, **kwargs, distributed = distributed)
base_decoder.train(was_training)
return out
trainable_unets = self.accelerator.unwrap_model(self.decoder).unets trainable_unets = self.accelerator.unwrap_model(self.decoder).unets
base_decoder.unets = self.unets # swap in exponential moving averaged unets for sampling base_decoder.unets = self.unets # swap in exponential moving averaged unets for sampling
@@ -687,6 +693,7 @@ class DecoderTrainer(nn.Module):
for ema in self.ema_unets: for ema in self.ema_unets:
ema.restore_ema_model_device() ema.restore_ema_model_device()
base_decoder.train(was_training)
return output return output
@torch.no_grad() @torch.no_grad()

2
dalle2_pytorch/version.py
View File

@@ -1 +1 @@
__version__ = '0.23.6' __version__ = '0.23.7'