{
    "training": {
        "seed": 777,
        "logging_steps": 100,
        "eval_steps": 500,
        "gradient_accumulation_steps": 2,
        "gradient_clip_value": 1.0,
        "learning_rate": 5e-05,
        "betas": [0.9, 0.95],
        "warmup_ratio": 0.03,
        "batch_size": 4,
        "weight_decay": 0.1,
        "precision": "bf16",
        "strategy": "ddp",
        "gradient_checkpointing": true,
        "num_workers": 2,
        "max_tokens_per_batch": 16384,
        "min_lr_ratio": 0.1
    },
    "modeling": {
        "parameters": {
            "codebook_size": 65536,
            "max_seq_len": 2048,
            "model_name": "meta-llama/Llama-3.2-1B-Instruct",
            "enable_text_normalization": true
        }
    },
    "checkpointing": {
        "save_steps": 500,
        "collect_health_stats": true,
        "save_intermediate_generations": true,
        "validation_type": "prompt_continuation",
        "only_load_model_weights": true,
        "keep_only_last_n_checkpoints": 5,
        "codec_encoder_path": "/path/to/codec_encoder.pt",
        "codec_decoder_path": "/path/to/codec_decoder.pt"
    },
    "train_weighted_datasets": {
        "/path/to/your/vectorized_dataset": 1.0
    },
    "val_weighted_datasets": {
        "/path/to/your/vectorized_dataset": 1.0
    },
    "dataset": {
        "allowed_languages": [],
        "min_dnsmos_score": 3.0,
        "min_sample_rate": 16000,
        "enable_rlhf_training": false,
        "min_audio_duration": 0.5
    }
}