{
    "model_name": "Transformer XL LM",
    "model_type": "LM",

    "lm_params": 
    {
        "arch": "Transformer",
        "num_blocks": 12,
        "dim_model": 768,
        "ff_ratio": 4,
        "num_heads": 12,
        "vocab_size": 1000,
        "relative_pos_enc": true,
        "max_pos_encoding": 2048,
        "Pdrop": 0.1
    },
    
    "tokenizer_params":
    {
        "tokenizer_path": "datasets/LibriSpeech/LibriSpeech_bpe_1000.model",
        "vocab_type": "bpe",
        "vocab_size": 1000
    },

    "training_params":
    {
        "epochs": 100,
        "batch_size": 64,
        "accumulated_steps": 5,
        "mixed_precision": true,

        "optimizer": "Adam",
        "beta1": 0.9,
        "beta2": 0.95,
        "eps": 1e-8,
        "weight_decay": 0.0,

        "lr_schedule": "Cosine",
        "warmup_steps": 1000,
        "end_step": 300000,
        "lr_max": 0.0006,
        "lr_min": 0.00006,

        "train_label_max_length": 100,
        "eval_audio_max_length": null,
        "eval_label_max_length": null,

        "training_dataset": "LibriSpeechCorpus",
        "training_dataset_path": "datasets/librispeech-lm-norm.txt",

        "evaluation_dataset": "LibriSpeech",
        "evaluation_dataset_path": "datasets/LibriSpeech/",
        "lm_mode": true,

        "callback_path": "callbacks/LM-Transformer/"
    },

    "decoding_params":
    {
        "beam_size": 16,
        "tmp": 1
    }
}