{
    "project_name": "mms_tn_finetuning",
    "push_to_hub": true,
    "hub_model_id": "Arbi-Houssem/mms_tts_tun_Lang1.3",
    "report_to": ["wandb"],
    "overwrite_output_dir": true,
    "output_dir": "./tmp/vits_finetuned_tn3",

    "dataset_name": "Arbi-Houssem/datasetSTT-TTS",     
    "audio_column_name": "audio", 
    "text_column_name":"sentence",
    "train_split_name": "train",
    "eval_split_name": "validation",
    "override_speaker_embeddings": true,


    "full_generation_sample_text": "نحب شكون يعاوني بش نشري حاجات من عندكم",
    
    "max_duration_in_seconds": 30,
    "min_duration_in_seconds": 1.0,
    "max_tokens_length": 2000,

    "model_name_or_path": "Arbi-Houssem/mms_tts_tun_Lang1.3",


    "preprocessing_num_workers": 4,

    "do_train": true,
    "num_train_epochs": 650,
    "gradient_accumulation_steps": 1,
    "gradient_checkpointing": true,
    "per_device_train_batch_size": 8,
    "learning_rate": 2e-6,
    "adam_beta1": 0.8,
    "adam_beta2": 0.99,
    "warmup_ratio": 0.01,
    "group_by_length": false,

    "do_eval": true, 
    "eval_steps": 700,
    "per_device_eval_batch_size": 8,
    "max_eval_samples": 22, 
    "do_step_schedule_per_epoch": true,

    "weight_disc": 3,
    "weight_fmaps": 1,
    "weight_gen": 1,
    "weight_kl": 1.5,
    "weight_duration": 1,
    "weight_mel": 35,

    "fp16": true,
    "seed": 456
}