tuna2134/sbv2-v2
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

rollback resume

Browse Source
This commit is contained in:
Stardust·减
2023-09-04 22:13:57 +08:00
committed by GitHub
parent c6f627f350
commit 7761dadcc6
1 changed files with 9 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
train_ms.py
View File

@@ -155,34 +155,20 @@ def run(rank, n_gpus, hps):
net_d = DDP(net_d, device_ids=[rank], find_unused_parameters=True) net_d = DDP(net_d, device_ids=[rank], find_unused_parameters=True)
if net_dur_disc is not None: if net_dur_disc is not None:
net_dur_disc = DDP(net_dur_disc, device_ids=[rank], find_unused_parameters=True) net_dur_disc = DDP(net_dur_disc, device_ids=[rank], find_unused_parameters=True)
try:
if net_dur_disc is not None:
_, _, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "DUR_*.pth"), net_dur_disc, optim_dur_disc, skip_optimizer=True)
_, optim_g, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "G_*.pth"), net_g,
optim_g, skip_optimizer=True)
_, optim_d, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "D_*.pth"), net_d,
optim_d, skip_optimizer=True)
pretrain_dir = None epoch_str = max(epoch_str, 1)
if pretrain_dir is None: global_step = (epoch_str - 1) * len(train_loader)
try:
if net_dur_disc is not None:
_, optim_dur_disc, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "DUR_*.pth"), net_dur_disc, optim_dur_disc, skip_optimizer=not hps.resume)
_, optim_g, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "G_*.pth"), net_g,
optim_g, skip_optimizer=not hps.resume)
_, optim_d, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "D_*.pth"), net_d,
optim_d, skip_optimizer=not hps.resume)
else:
_, optim_g, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "G_*.pth"), net_g,
optim_g, skip_optimizer=not hps.resume)
_, optim_d, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "D_*.pth"), net_d,
optim_d, skip_optimizer=not hps.resume)
epoch_str = max(epoch_str, 1)
global_step = (epoch_str - 1) * len(train_loader)
except Exception as e: except Exception as e:
print(e) print(e)
epoch_str = 1 epoch_str = 1
global_step = 0 global_step = 0
else:
_, _, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(pretrain_dir, "G_*.pth"), net_g,
optim_g, True)
_, _, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(pretrain_dir, "D_*.pth"), net_d,
optim_d, True)
scheduler_g = torch.optim.lr_scheduler.ExponentialLR(optim_g, gamma=hps.train.lr_decay, last_epoch=epoch_str - 2) scheduler_g = torch.optim.lr_scheduler.ExponentialLR(optim_g, gamma=hps.train.lr_decay, last_epoch=epoch_str - 2)