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

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

GPT_SoVITS/s2_train_v3.py

@@ -77,6 +77,7 @@ 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"))
 
+    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",
@@ -166,6 +167,16 @@ def run(rank, n_gpus, hps):
     #     eps=hps.train.eps,
     # )
     if torch.cuda.is_available():
+        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:

GPT_SoVITS/s2_train_v3_lora.py

@@ -77,6 +77,7 @@ 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"))
 
+    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",
@@ -156,6 +157,15 @@ def run(rank, n_gpus, hps):
 
     def model2cuda(net_g, rank):
         if torch.cuda.is_available():
+            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)