{
    "model_name": "Efficient Conformer Transducer Small",
    "model_type": "Transducer",

    "encoder_params": 
    {
        "arch": "Conformer",
        "num_blocks": 15,
        "dim_model": [100, 140, 200],
        "ff_ratio": 4,
        "num_heads": 4,
        "kernel_size": 15,
        "Pdrop": 0.1,
        "conv_stride": 2,
        "att_stride": 1,
        "strided_blocks": [4, 9],
        "expand_blocks": [4, 9],
        "att_group_size": [3, 1, 1],

        "relative_pos_enc": true,
        "max_pos_encoding": 10000,

        "subsampling_module": "Conv2d",
        "subsampling_layers": 1,
        "subsampling_filters": [100],
        "subsampling_kernel_size": 3,
        "subsampling_norm": "batch",
        "subsampling_act": "swish",

        "sample_rate": 16000,
        "win_length_ms": 25,
        "hop_length_ms": 10,
        "n_fft": 512,
        "n_mels": 80,
        "normalize": false,
        "mean": -5.6501,
        "std": 4.2280,

        "spec_augment": true,
        "mF": 2,
        "F": 27,
        "mT": 10,
        "pS": 0.05
    },

    "decoder_params":
    {
        "arch": "RNN",
        "num_layers": 1,
        "dim_model": 320,
        "vocab_size": 1000
    },

    "joint_params":
    {
        "joint_mode": "sum",
        "dim_model": 320,
        "act": "tanh"
    },
    
    "tokenizer_params":
    {
        "tokenizer_path": "datasets/LibriSpeech/LibriSpeech_bpe_1000.model",
        "vocab_type": "bpe",
        "vocab_size": 1000
    },

    "training_params":
    {
        "epochs": 250,
        "batch_size": 16,
        "accumulated_steps": 4,
        "mixed_precision": true,

        "optimizer": "Adam",
        "beta1": 0.9,
        "beta2": 0.98,
        "eps": 1e-9,
        "weight_decay": 1e-6,

        "lr_schedule": "Transformer",
        "schedule_dim": 200,
        "warmup_steps": 10000,
        "K": 5,

        "vn_start_step": 20000,
        "vn_std": 0.075,

        "train_audio_max_length": 256000,
        "train_label_max_length": 90,
        "eval_audio_max_length": null,
        "eval_label_max_length": null,

        "training_dataset": "LibriSpeech",
        "training_dataset_path": "datasets/LibriSpeech/",

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

        "callback_path": "callbacks/EfficientConformerTransducerSmall/"
    },

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

        "ngram_path": "callbacks/ngram/6gram_1000.arpa",
        "ngram_alpha": 0.3,
        "ngram_beta": 1,

        "lm_config": "configs/LM-Transformer.json",
        "lm_weight": 1,
        "lm_tmp": 1
    }
}