singing.yaml

# @package _global_

# To pick dataset path execute with: ++datamodule.dataset.path=/your_wav_root/

sampling_rate: 24000
length: 131072
channels: 1
log_every_n_steps: 3000
overlap: 65536

model:
  _target_: main.module_vckt.Model
  lr: 1e-4
  lr_beta1: 0.95
  lr_beta2: 0.999
  lr_eps: 1e-6
  lr_weight_decay: 1e-3
  # ema_beta: 0.9999
  # ema_power: 0.7

  model:
    _target_: audio_diffusion_pytorch.AudioDiffusionModel
    in_channels: ${channels}
    channels: 256
    patch_size: 12
    resnet_groups: 8
    kernel_multiplier_downsample: 2
    multipliers: [ 1, 2, 4, 4, 4, 4, 4 ]
    factors: [ 4, 4, 4, 2, 2, 2 ]
    num_blocks: [ 2, 2, 2, 2, 2, 2 ]
    attentions: [ 0, 0, 0, 1, 1, 1, 1 ]
    attention_heads: 8
    attention_features: 128
    attention_multiplier: 2
    use_nearest_upsample: False
    use_skip_scale: True
    diffusion_type: k
    diffusion_sigma_distribution:
      _target_: audio_diffusion_pytorch.LogNormalDistribution
      mean: -3.0
      std: 1.0
    diffusion_sigma_data: 0.2


datamodule:
  _target_: main.module_vckt.Datamodule
  dataset:
    _target_: torch.utils.data.ConcatDataset
    datasets:
      - _target_: datasets.wav.WAVDataset
        data_dir: /import/c4dm-datasets-ext/m4singer-24k/
        segment: ${length}
        overlap: ${overlap}
        mono: True
        keepdim: True
      - _target_: datasets.wav.WAVDataset
        data_dir: /import/c4dm-datasets-ext/ycy_artefacts/OpenCPOP-24k/
        segment: ${length}
        overlap: ${overlap}
        mono: True
        keepdim: True
      - _target_: datasets.wav.WAVDataset
        data_dir: /import/c4dm-datasets-ext/ycy_artefacts/VocalSet-24k/
        segment: ${length}
        overlap: ${overlap}
        mono: True
        keepdim: True
      - _target_: datasets.wav.WAVDataset
        data_dir: /import/c4dm-datasets-ext/ycy_artefacts/OpenSinger-24k/
        segment: ${length}
        overlap: ${overlap}
        mono: True
        keepdim: True
      - _target_: datasets.wav.WAVDataset
        data_dir: /import/c4dm-datasets-ext/jvs_music_ver1/
        segment: ${length}
        overlap: ${overlap}
        mono: True
        keepdim: True
      - _target_: datasets.wav.WAVDataset
        data_dir: /import/c4dm-datasets-ext/ycy_artefacts/CSD-24k/
        segment: ${length}
        overlap: ${overlap}
        mono: True
        keepdim: True
      - _target_: datasets.wav.WAVDataset
        data_dir: /import/c4dm-datasets-ext/ycy_artefacts/NUS-24k/
        segment: ${length}
        overlap: ${overlap}
        mono: True
        keepdim: True
      - _target_: datasets.wav.WAVDataset
        data_dir: /import/c4dm-datasets-ext/ycy_artefacts/PJS-24k/
        segment: ${length}
        overlap: ${overlap}
        mono: True
        keepdim: True


  collate_fn: null

  val_split: 0.01
  batch_size: 32
  num_workers: 8
  pin_memory: True

callbacks:
  rich_progress_bar:
    _target_: pytorch_lightning.callbacks.RichProgressBar

  model_checkpoint:
    _target_: pytorch_lightning.callbacks.ModelCheckpoint
    monitor: "valid_loss"   # name of the logged metric which determines when model is improving
    save_top_k: 1           # save k best models (determined by above metric)
    save_last: True         # additionaly always save model from last epoch
    mode: "min"             # can be "max" or "min"
    verbose: False
    dirpath: ${logs_dir}/ckpts/${now:%Y-%m-%d-%H-%M-%S}
    filename: '{epoch:02d}-{valid_loss:.3f}'

  model_summary:
    _target_: pytorch_lightning.callbacks.RichModelSummary
    max_depth: 2

  audio_samples_logger:
    _target_: main.module_vckt.SampleLogger
    num_items: 4
    channels: ${channels}
    sampling_rate: ${sampling_rate}
    length: ${length}
    sampling_steps: [100]
    # use_ema_model: True
    diffusion_sampler:
      _target_: audio_diffusion_pytorch.ADPM2Sampler
      rho: 1.0
    diffusion_schedule:
      _target_: audio_diffusion_pytorch.KarrasSchedule
      sigma_min: 0.0001
      sigma_max: 5.0
      rho: 9.0

loggers:
  wandb:
    _target_: pytorch_lightning.loggers.wandb.WandbLogger
    project: ${oc.env:WANDB_PROJECT}
    entity: ${oc.env:WANDB_ENTITY}
    # offline: False  # set True to store all logs only locally
    job_type: "train"
    group: ""
    save_dir: ${logs_dir}

trainer:
  _target_: pytorch_lightning.Trainer
  gpus: -1 # Set `1` to train on GPU, `0` to train on CPU only, and `-1` to train on all GPUs, default `0`
  precision: 32 # Precision used for tensors, default `32`
  accelerator: null # `ddp` GPUs train individually and sync gradients, default `None`
  min_epochs: 0
  max_epochs: -1
  enable_model_summary: False
  log_every_n_steps: 1 # Logs metrics every N batches
  check_val_every_n_epoch: null
  val_check_interval: ${log_every_n_steps}