Skip to content

SOTA Open Source TTS

update some deault config value (#19)

faceairfa62856e2534
diff --git a/fish_speech/configs/vqgan_finetune.yaml b/fish_speech/configs/vqgan_finetune.yaml
index 6d25bb1..aa5ddaa 100644
--- a/fish_speech/configs/vqgan_finetune.yaml
+++ b/fish_speech/configs/vqgan_finetune.yaml
@@ -9,7 +9,7 @@ resume_weights_only: true
# Lightning Trainer
trainer:
accelerator: gpu
- devices: 4
+ devices: auto
strategy: ddp_find_unused_parameters_true
precision: 32
max_steps: 100000
diff --git a/fish_speech/configs/vqgan_pretrain.yaml b/fish_speech/configs/vqgan_pretrain.yaml
index b01acfe..7dd3613 100644
--- a/fish_speech/configs/vqgan_pretrain.yaml
+++ b/fish_speech/configs/vqgan_pretrain.yaml
@@ -7,7 +7,7 @@ project: vqgan
# Lightning Trainer
trainer:
accelerator: gpu
- devices: 4
+ devices: auto
strategy: ddp_find_unused_parameters_true
precision: 32
max_steps: 1_000_000
diff --git a/pyproject.toml b/pyproject.toml
index 8317baf..23a9537 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -35,7 +35,8 @@ dependencies = [
"tensorboard",
"grpcio>=1.58.0",
"kui>=1.6.0",
- "zibai-server>=0.9.0"
+ "zibai-server>=0.9.0",
+ "loguru",
]
[build-system]
diff --git a/tools/vqgan/inference.py b/tools/vqgan/inference.py
index f81b20c..d0a222f 100644
--- a/tools/vqgan/inference.py
+++ b/tools/vqgan/inference.py
@@ -33,9 +33,7 @@ OmegaConf.register_new_resolver("eval", eval)
"--output-path", "-o", default="fake.wav", type=click.Path(path_type=Path)
)
@click.option("--config-name", "-cfg", default="vqgan_pretrain")
- "--checkpoint-path", "-ckpt", default="checkpoints/vqgan/step_000380000_wo.ckpt"
-)
[email protected]("--checkpoint-path", "-ckpt", default="checkpoints/vqgan-v1.pth")
def main(input_path, output_path, config_name, checkpoint_path):
with initialize(version_base="1.3", config_path="../../fish_speech/configs"):
cfg = compose(config_name=config_name)