config.yaml

hydra:
  run:
    dir: ckpts/${datasets.name}_{model.tokenizer}

datasets:
  name: ${datasets.name}
  batch_size_per_gpu: 8  
  batch_size_type: sample  
  max_samples: 1 
  num_workers: 16

optim:
  epochs: 250  
  learning_rate: 7.5e-5
  num_warmup_updates: 20000  
  grad_accumulation_steps: 1  
  max_grad_norm: 1.0  
  bnb_optimizer: False  

model:
  name: SilmaTTS_V1_Small
  tokenizer: char
  tokenizer_path: null  
  backbone: DiT
  arch:
    dim: 768
    depth: 18
    heads: 12
    ff_mult: 2
    text_dim: 512
    text_mask_padding: True
    conv_layers: 4
    pe_attn_head: null
    attn_backend: torch  
    attn_mask_enabled: False
    checkpoint_activations: False  
  mel_spec:
    target_sample_rate: 24000
    n_mel_channels: 100
    hop_length: 256
    win_length: 1024
    n_fft: 1024
    mel_spec_type: vocos  
  vocoder:
    is_local: False  
    local_path: null  

ckpts:
  logger: wandb  
  log_samples: True 
  save_per_updates: 1000  
  keep_last_n_checkpoints: 100  
  last_per_updates: 1000  
  save_dir: ckpts/${datasets.name}_{model.tokenizer}