Training is now compleletly unusable

My training has a strange behavior, it can only produce index file, but not model file.
When hiting train model button,

OS: Win10
GPU: NVIDIA 4080S

It prints this

Traceback (most recent call last):
  File "multiprocessing\process.py", line 315, in _bootstrap
  File "multiprocessing\process.py", line 108, in run
  File "C:\RVC\infer\modules\train\train.py", line 233, in run
    net_g.module.load_state_dict(
  File "C:\RVC\runtime\lib\site-packages\torch\nn\modules\module.py", line 2041, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for SynthesizerTrnMs768NSFsid:
        size mismatch for dec.ups.0.weight_v: copying a param with shape torch.Size([512, 256, 24]) from checkpoint, the shape in current model is torch.Size([512, 256, 16]).
        size mismatch for dec.ups.1.weight_v: copying a param with shape torch.Size([256, 128, 20]) from checkpoint, the shape in current model is torch.Size([256, 128, 16]).
2024-09-28 20:26:31 | INFO | __main__ | Use gpus: 0
2024-09-28 20:26:31 | INFO | __main__ | "runtime\python.exe" infer/modules/train/train.py -e "test_project" -sr 40k -f0 1 -bs 8 -g 0 -te 20 -se 5 -pg assets/pretrained_v2/f0G48k.pth -pd assets/pretrained_v2/f0D48k.pth -l 0 -c 1 -sw 0 -v v2
INFO:test_project:{'data': {'filter_length': 2048, 'hop_length': 400, 'max_wav_value': 32768.0, 'mel_fmax': None, 'mel_fmin': 0.0, 'n_mel_channels': 125, 'sampling_rate': 40000, 'win_length': 2048, 'training_files': './logs\\test_project/filelist.txt'}, 'model': {'filter_channels': 768, 'gin_channels': 256, 'hidden_channels': 192, 'inter_channels': 192, 'kernel_size': 3, 'n_heads': 2, 'n_layers': 6, 'p_dropout': 0, 'resblock': '1', 'resblock_dilation_sizes': [[1, 3, 5], [1, 3, 5], [1, 3, 5]], 'resblock_kernel_sizes': [3, 7, 11], 'spk_embed_dim': 109, 'upsample_initial_channel': 512, 'upsample_kernel_sizes': [16, 16, 4, 4], 'upsample_rates': [10, 10, 2, 2], 'use_spectral_norm': False}, 'train': {'batch_size': 8, 'betas': [0.8, 0.99], 'c_kl': 1.0, 'c_mel': 45, 'epochs': 20000, 'eps': 1e-09, 'fp16_run': True, 'init_lr_ratio': 1, 'learning_rate': 0.0001, 'log_interval': 200, 'lr_decay': 0.999875, 'seed': 1234, 'segment_size': 12800, 'warmup_epochs': 0}, 'model_dir': './logs\\test_project', 'experiment_dir': './logs\\test_project', 'save_every_epoch': 5, 'name': 'test_project', 'total_epoch': 20, 'pretrainG': 'assets/pretrained_v2/f0G48k.pth', 'pretrainD': 'assets/pretrained_v2/f0D48k.pth', 'version': 'v2', 'gpus': '0', 'sample_rate': '40k', 'if_f0': 1, 'if_latest': 0, 'save_every_weights': '0', 'if_cache_data_in_gpu': 1}
[W C:\actions-runner\_work\pytorch\pytorch\builder\windows\pytorch\torch\csrc\distributed\c10d\socket.cpp:601] [c10d] The client socket has failed to connect to [mytestserver.com]:37197 (system error: 10049 - 在其上 下文中，该请求的地址无效。).
[W C:\actions-runner\_work\pytorch\pytorch\builder\windows\pytorch\torch\csrc\distributed\c10d\socket.cpp:601] [c10d] The client socket has failed to connect to [mytestserver.com]:37197 (system error: 10049 - 在其上 下文中，该请求的地址无效。).

It is rediculous, it even says it is refering to
"C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\torch\csrc\distributed\c10d\socket.cpp"
I even does not have a folder called actions-runner at C drive
And rediculously, the address it is connecting to is my custom sample server.

Start with infer/modules/train/train.py around line 233 and reproduce the Windows training command shown in the report. Compare the pretrained checkpoint shapes with the logged model configuration, then investigate the PyTorch distributed socket warnings. Done means training completes without the state_dict size mismatch and produces the expected model file.