bacow/GPT-SoVITS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Refactor: Format Code with Ruff and Update Deprecated G2PW Link (#2255)

Browse Source 
* ruff check --fix

* ruff format --line-length 120 --target-version py39

* Change the link for G2PW Model

* update pytorch version and colab
This commit is contained in:
XXXXRT666
2025-04-07 09:42:47 +01:00
committed by GitHub
parent 9da7e17efe
commit 53cac93589
132 changed files with 8185 additions and 6648 deletions
Download Patch File Download Diff File Expand all files Collapse all files

66
GPT_SoVITS/BigVGAN/bigvgan.py
View File

@@ -50,7 +50,7 @@ class AMPBlock1(torch.nn.Module):
activation: str = None,
):
super().__init__()
self.h = h
self.convs1 = nn.ModuleList(
@@ -87,9 +87,7 @@ class AMPBlock1(torch.nn.Module):
)
self.convs2.apply(init_weights)
self.num_layers = len(self.convs1) + len(
self.convs2
) # Total number of conv layers
self.num_layers = len(self.convs1) + len(self.convs2) # Total number of conv layers
# Select which Activation1d, lazy-load cuda version to ensure backward compatibility
if self.h.get("use_cuda_kernel", False):
@@ -105,22 +103,14 @@ class AMPBlock1(torch.nn.Module):
if activation == "snake":
self.activations = nn.ModuleList(
[
Activation1d(
activation=activations.Snake(
channels, alpha_logscale=h.snake_logscale
)
)
Activation1d(activation=activations.Snake(channels, alpha_logscale=h.snake_logscale))
for _ in range(self.num_layers)
]
)
elif activation == "snakebeta":
self.activations = nn.ModuleList(
[
Activation1d(
activation=activations.SnakeBeta(
channels, alpha_logscale=h.snake_logscale
)
)
Activation1d(activation=activations.SnakeBeta(channels, alpha_logscale=h.snake_logscale))
for _ in range(self.num_layers)
]
)
@@ -169,7 +159,7 @@ class AMPBlock2(torch.nn.Module):
activation: str = None,
):
super().__init__()
self.h = h
self.convs = nn.ModuleList(
@@ -205,22 +195,14 @@ class AMPBlock2(torch.nn.Module):
if activation == "snake":
self.activations = nn.ModuleList(
[
Activation1d(
activation=activations.Snake(
channels, alpha_logscale=h.snake_logscale
)
)
Activation1d(activation=activations.Snake(channels, alpha_logscale=h.snake_logscale))
for _ in range(self.num_layers)
]
)
elif activation == "snakebeta":
self.activations = nn.ModuleList(
[
Activation1d(
activation=activations.SnakeBeta(
channels, alpha_logscale=h.snake_logscale
)
)
Activation1d(activation=activations.SnakeBeta(channels, alpha_logscale=h.snake_logscale))
for _ in range(self.num_layers)
]
)
@@ -283,9 +265,7 @@ class BigVGAN(
self.num_upsamples = len(h.upsample_rates)
# Pre-conv
self.conv_pre = weight_norm(
Conv1d(h.num_mels, h.upsample_initial_channel, 7, 1, padding=3)
)
self.conv_pre = weight_norm(Conv1d(h.num_mels, h.upsample_initial_channel, 7, 1, padding=3))
# Define which AMPBlock to use. BigVGAN uses AMPBlock1 as default
if h.resblock == "1":
@@ -293,9 +273,7 @@ class BigVGAN(
elif h.resblock == "2":
resblock_class = AMPBlock2
else:
raise ValueError(
f"Incorrect resblock class specified in hyperparameters. Got {h.resblock}"
)
raise ValueError(f"Incorrect resblock class specified in hyperparameters. Got {h.resblock}")
# Transposed conv-based upsamplers. does not apply anti-aliasing
self.ups = nn.ModuleList()
@@ -320,22 +298,14 @@ class BigVGAN(
self.resblocks = nn.ModuleList()
for i in range(len(self.ups)):
ch = h.upsample_initial_channel // (2 ** (i + 1))
for j, (k, d) in enumerate(
zip(h.resblock_kernel_sizes, h.resblock_dilation_sizes)
):
self.resblocks.append(
resblock_class(h, ch, k, d, activation=h.activation)
)
for j, (k, d) in enumerate(zip(h.resblock_kernel_sizes, h.resblock_dilation_sizes)):
self.resblocks.append(resblock_class(h, ch, k, d, activation=h.activation))
# Post-conv
activation_post = (
activations.Snake(ch, alpha_logscale=h.snake_logscale)
if h.activation == "snake"
else (
activations.SnakeBeta(ch, alpha_logscale=h.snake_logscale)
if h.activation == "snakebeta"
else None
)
else (activations.SnakeBeta(ch, alpha_logscale=h.snake_logscale) if h.activation == "snakebeta" else None)
)
if activation_post is None:
raise NotImplementedError(
@@ -346,9 +316,7 @@ class BigVGAN(
# Whether to use bias for the final conv_post. Default to True for backward compatibility
self.use_bias_at_final = h.get("use_bias_at_final", True)
self.conv_post = weight_norm(
Conv1d(ch, 1, 7, 1, padding=3, bias=self.use_bias_at_final)
)
self.conv_post = weight_norm(Conv1d(ch, 1, 7, 1, padding=3, bias=self.use_bias_at_final))
# Weight initialization
for i in range(len(self.ups)):
@@ -451,13 +419,13 @@ class BigVGAN(
# instantiate BigVGAN using h
if use_cuda_kernel:
print(
f"[WARNING] You have specified use_cuda_kernel=True during BigVGAN.from_pretrained(). Only inference is supported (training is not implemented)!"
"[WARNING] You have specified use_cuda_kernel=True during BigVGAN.from_pretrained(). Only inference is supported (training is not implemented)!"
)
print(
f"[WARNING] You need nvcc and ninja installed in your system that matches your PyTorch build is using to build the kernel. If not, the model will fail to initialize or generate incorrect waveform!"
"[WARNING] You need nvcc and ninja installed in your system that matches your PyTorch build is using to build the kernel. If not, the model will fail to initialize or generate incorrect waveform!"
)
print(
f"[WARNING] For detail, see the official GitHub repository: https://github.com/NVIDIA/BigVGAN?tab=readme-ov-file#using-custom-cuda-kernel-for-synthesis"
"[WARNING] For detail, see the official GitHub repository: https://github.com/NVIDIA/BigVGAN?tab=readme-ov-file#using-custom-cuda-kernel-for-synthesis"
)
model = cls(h, use_cuda_kernel=use_cuda_kernel)
@@ -485,7 +453,7 @@ class BigVGAN(
model.load_state_dict(checkpoint_dict["generator"])
except RuntimeError:
print(
f"[INFO] the pretrained checkpoint does not contain weight norm. Loading the checkpoint after removing weight norm!"
"[INFO] the pretrained checkpoint does not contain weight norm. Loading the checkpoint after removing weight norm!"
)
model.remove_weight_norm()
model.load_state_dict(checkpoint_dict["generator"])