config_s2a.json

{
  "saving_path": "/home/ubuntu/experiments/a2s_mls",
  "resume_checkpoint": null,
  "vocoder_type": "SPEECHTOKENIZER",
  "vocoder_config_path": null,
  "vocoder_ckpt_path": null,
  "metapath": [
    "/var/data_mls/train.json"
  ],
  "val_metapath": [
    "/var/data_mls/test.json"
  ],
  "pretrained_path": null,
  "speaker_embedding_dir": null,
  "sampledir": "/home/ubuntu/experiments/a2s_mls",
  "lr": 0.0005,
  "batch_size": 100.0,
  "train_bucket_size": 8192,
  "training_step": 800000,
  "optim_flat_percent": 0.0,
  "warmup_step": 10000,
  "adam_beta1": 0.9,
  "adam_beta2": 0.98,
  "ffd_size": 1024,
  "hidden_size": 1024,
  "enc_nlayers": 8,
  "dec_nlayers": 6,
  "nheads": 8,
  "dropout": 0.1,
  "depthwise_conv_kernel_size": 5,
  "aligner_softmax_temp": 1.0,
  "layer_norm_eps": 1e-05,
  "use_sem_tokens": true,
  "use_spkr_emb": false,
  "use_text_emb": false,
  "fairseq": false,
  "only_inference": false,
  "speaker_embed_dropout": 0.05,
  "label_smoothing": 0.0,
  "val_check_interval": 1,
  "max_dataset_samples": -1,
  "check_val_every_n_epoch": 1,
  "precision": "bf16",
  "nworkers": 12,
  "distributed": true,
  "accelerator": "gpu",
  "version": null,
  "accumulate_grad_batches": 1,
  "sagemaker": false,
  "use_repetition_token": false,
  "use_repetition_gating": false,
  "repetition_penalty": 1.0,
  "sampling_temperature": 1.0,
  "top_k": -1,
  "min_top_k": 3,
  "top_p": 0.8,
  "sample_num": 4,
  "length_penalty_max_length": 150,
  "length_penalty_max_prob": 0.95,
  "max_input_length": 2048,
  "max_output_length": 2000,
  "phone_context_window": 3,
  "sample_rate": 16000,
  "n_codes": 1024,
  "n_cluster_groups": 7,
  "first_n_lvls": 7,
  "use_pretrained_ckpt_cfg": false,
  "n_semantic_codes": 1024
}