base_model:
  attn_mask_enabled: false
  conv_layers: 4
  depth: 22
  dim: 1024
  ff_mult: 2
  heads: 16
  pe_attn_head: 1
  text_dim: 512
  text_mask_padding: false
ckpts:
  keep_last_n_checkpoints: 3
  last_per_updates: 1000
  log_samples: false
  logger: wandb
  save_dir: ckpts/uk_regional
  save_per_updates: 5000
dataset:
  audio_folder: ''
  dataset_type: CustomDatasetPath
  hop_length: 256
  mel_fmax: 8000
  mel_fmin: 0
  n_mel_channels: 100
  name: /home/ubuntu/regional_voice_engine/datasets/uk_dialects/arrow_data
  sample_rate: 24000
  tokenizer: char
  vocab_file: /home/ubuntu/regional_voice_engine/datasets/uk_dialects/arrow_data/vocab.txt
datasets:
  batch_size_per_gpu: 4
  batch_size_type: sample
  max_samples: 16
  name: /home/ubuntu/regional_voice_engine/datasets/uk_dialects/arrow_data
  num_workers: 2
generated_tracker_name:
  name: F5-TTS-UK-Regional
hydra:
  run:
    dir: ckpts/uk_regional/${now:%Y-%m-%d}/${now:%H-%M-%S}
model:
  arch:
    attn_mask_enabled: false
    conv_layers: 4
    depth: 22
    dim: 1024
    ff_mult: 2
    heads: 16
    pe_attn_head: 1
    text_dim: 512
    text_mask_padding: false
  backbone: DiT
  mel_spec:
    hop_length: 256
    mel_spec_type: vocos
    n_fft: 1024
    n_mel_channels: 100
    target_sample_rate: 24000
    win_length: 1024
  name: F5TTS_UK
  tokenizer: custom
  tokenizer_path: /home/ubuntu/regional_voice_engine/datasets/uk_dialects/arrow_data/vocab.txt
  vocoder:
    is_local: false
    local_path: null
optim:
  bnb_optimizer: false
  epochs: 10000
  grad_accumulation_steps: 4
  learning_rate: 7.5e-05
  max_grad_norm: 1.0
  num_warmup_updates: 5000