blackteay/GPT-SoVITS
1
0
Fork 0
mirror of https://github.com/RVC-Boss/GPT-SoVITS.git synced 2026-05-21 02:22:42 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: apply same Windows single-GPU gloo bypass to s2_train_v3 and s2_train_v3_lora

Browse Source 
Extend the fix to v3 and LoRA training scripts:
- s2_train_v3.py: skip dist.init_process_group() + DummyDDP for Windows single-GPU
- s2_train_v3_lora.py: same fix applied to LoRA fine-tuning script
This commit is contained in:
fanfan-love-meatmeat 2026-03-05 11:44:25 +08:00
parent 832e5b6160
commit 4820d5a101
2 changed files with 36 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
GPT_SoVITS/s2_train_v3.py
View File

@ -77,12 +77,13 @@ def run(rank, n_gpus, hps):
writer = SummaryWriter(log_dir=hps.s2_ckpt_dir)
writer_eval = SummaryWriter(log_dir=os.path.join(hps.s2_ckpt_dir, "eval"))
dist.init_process_group(
backend="gloo" if os.name == "nt" or not torch.cuda.is_available() else "nccl",
init_method="env://?use_libuv=False",
world_size=n_gpus,
rank=rank,
)
if not (os.name == "nt" and n_gpus == 1):
dist.init_process_group(
backend="gloo" if os.name == "nt" or not torch.cuda.is_available() else "nccl",
init_method="env://?use_libuv=False",
world_size=n_gpus,
rank=rank,
)
torch.manual_seed(hps.train.seed)
if torch.cuda.is_available():
torch.cuda.set_device(rank)
@ -166,8 +167,18 @@ def run(rank, n_gpus, hps):
# eps=hps.train.eps,
# )
if torch.cuda.is_available():
net_g = DDP(net_g, device_ids=[rank], find_unused_parameters=True)
# net_d = DDP(net_d, device_ids=[rank], find_unused_parameters=True)
if os.name == "nt" and n_gpus == 1:
class DummyDDP(torch.nn.Module):
def __init__(self, module):
super().__init__()
self.module = module
def forward(self, *args, **kwargs):
return self.module(*args, **kwargs)
net_g = DummyDDP(net_g)
# net_d = DummyDDP(net_d)
else:
net_g = DDP(net_g, device_ids=[rank], find_unused_parameters=True)
# net_d = DDP(net_d, device_ids=[rank], find_unused_parameters=True)
else:
net_g = net_g.to(device)
# net_d = net_d.to(device)

24
GPT_SoVITS/s2_train_v3_lora.py
View File

@ -77,12 +77,13 @@ def run(rank, n_gpus, hps):
writer = SummaryWriter(log_dir=hps.s2_ckpt_dir)
writer_eval = SummaryWriter(log_dir=os.path.join(hps.s2_ckpt_dir, "eval"))
dist.init_process_group(
backend="gloo" if os.name == "nt" or not torch.cuda.is_available() else "nccl",
init_method="env://?use_libuv=False",
world_size=n_gpus,
rank=rank,
)
if not (os.name == "nt" and n_gpus == 1):
dist.init_process_group(
backend="gloo" if os.name == "nt" or not torch.cuda.is_available() else "nccl",
init_method="env://?use_libuv=False",
world_size=n_gpus,
rank=rank,
)
torch.manual_seed(hps.train.seed)
if torch.cuda.is_available():
torch.cuda.set_device(rank)
@ -156,7 +157,16 @@ def run(rank, n_gpus, hps):
def model2cuda(net_g, rank):
if torch.cuda.is_available():
net_g = DDP(net_g.cuda(rank), device_ids=[rank], find_unused_parameters=True)
if os.name == "nt" and n_gpus == 1:
class DummyDDP(torch.nn.Module):
def __init__(self, module):
super().__init__()
self.module = module
def forward(self, *args, **kwargs):
return self.module(*args, **kwargs)
net_g = DummyDDP(net_g.cuda(rank))
else:
net_g = DDP(net_g.cuda(rank), device_ids=[rank], find_unused_parameters=True)
else:
net_g = net_g.to(device)
return net_g