training_config.json

{
  "dataset_name": "amanuelbyte/acl-voice-cloning-fr-cleaned",
  "dataset_split": "train",
  "sample_fraction": 1.0,
  "cache_dir": "./data_cache",
  "audio_data_dir": "./audio_data",
  "native_sr": 16000,
  "max_audio_duration_sec": 20.0,
  "min_audio_duration_sec": 1.0,
  "ref_audio_max_duration_sec": 15.0,
  "filter_min_snr_db": 15.0,
  "filter_max_silence_ratio": 0.5,
  "filter_min_text_len": 5,
  "filter_max_text_len": 500,
  "output_dir": "chatterbox_fr_finetuned",
  "num_epochs": 5,
  "batch_size": 4,
  "gradient_accumulation_steps": 4,
  "learning_rate": 3e-05,
  "min_learning_rate": 3e-06,
  "max_grad_norm": 1.0,
  "warmup_ratio": 0.03,
  "fp16": true,
  "compile_model": false,
  "seed": 42,
  "weight_decay": 0.01,
  "use_lora": true,
  "lora_rank": 16,
  "lora_alpha": 32.0,
  "lora_dropout": 0.05,
  "target_modules": [
    "q_proj",
    "k_proj",
    "v_proj",
    "o_proj"
  ],
  "freeze_ve": true,
  "freeze_s3gen": true,
  "val_split_ratio": 0.1,
  "eval_every_ratio": 0.1,
  "save_every_ratio": 0.2,
  "log_every_n_steps": 5,
  "test_every_ratio": 0.25,
  "num_test_samples": 3,
  "patience": 8,
  "language_id": "fr",
  "cloning_mode": "cross_lingual",
  "hf_token": "",
  "hf_username": "amanuelbyte",
  "hf_repo_name": "chatterbox-fr-lora",
  "hf_repo_id": null,
  "skip_dataset_upload": true
}