Skip to content

SOTA Open Source TTS

Rename config & smaller schedule

Lengyue938eca21a67b
diff --git a/speech_lm/configs/pretrain.yaml b/speech_lm/configs/llama_pretrain.yaml
similarity index 100%
rename from speech_lm/configs/pretrain.yaml
rename to speech_lm/configs/llama_pretrain.yaml
diff --git a/speech_lm/configs/whisper_vq.yaml b/speech_lm/configs/whisper_vq.yaml
index e9cf992..c86ff79 100644
--- a/speech_lm/configs/whisper_vq.yaml
+++ b/speech_lm/configs/whisper_vq.yaml
@@ -11,7 +11,6 @@ trainer:
accelerator: gpu
strategy:
_target_: lightning.fabric.strategies.DDPStrategy
- find_unused_parameters: true
static_graph: true
devices: auto
@@ -40,31 +39,29 @@ model:
schedule:
batch_size: 64
micro_batch_size: 64
- max_steps: 1000000
+ max_steps: 10000
save_interval: 2000
gradient_accumulation_steps: "${eval: ${schedule.batch_size} // ${schedule.micro_batch_size}}"
clip_grad_norm: 2.0
-
-train_dataset:
- _target_: speech_lm.datasets.whisper_vq.WhisperVQDataset
- filelist: filelists/whisper-vq.train.train.filelist
-
-valid_dataset:
- _target_: speech_lm.datasets.whisper_vq.WhisperVQDataset
- filelist: filelists/whisper-vq.train.test.filelist
+ log_interval: 10
+ eval_interval: 2000
train_dataloader:
_target_: torch.utils.data.DataLoader
- dataset: ${dataset}
+ dataset:
+ _target_: speech_lm.datasets.whisper_vq.WhisperVQDataset
+ filelist: filelists/whisper-vq.train.train.filelist
batch_size: ${schedule.micro_batch_size}
- num_workers: 4
+ num_workers: 8
collate_fn:
_target_: speech_lm.datasets.whisper_vq.WhisperVQCollator
valid_dataloader:
_target_: torch.utils.data.DataLoader
- dataset: ${dataset}
- batch_size: ${schedule.micro_batch_size}
+ dataset:
+ _target_: speech_lm.datasets.whisper_vq.WhisperVQDataset
+ filelist: filelists/whisper-vq.train.test.filelist
+ batch_size: 32
num_workers: 4
collate_fn:
_target_: speech_lm.datasets.whisper_vq.WhisperVQCollator
diff --git a/speech_lm/train.py b/speech_lm/train.py
index cb11cf1..f2c3982 100644
--- a/speech_lm/train.py
+++ b/speech_lm/train.py
@@ -211,7 +211,7 @@ def train(
last_batch_time = time.time()
[email protected](version_base="1.3", config_path="./configs", config_name="pretrain.yaml")
[email protected](version_base="1.3", config_path="./configs", config_name="llama_pretrain.yaml")
def main(cfg: DictConfig):
log.info(f"Config: \n{OmegaConf.to_yaml(cfg)}")