SinaAhmadi/ScriptNormalization

1name: "Sorani-Arabic_1"2 3data:4    train: "datasets/Sorani-Arabic/1/train"5    dev: "datasets/Sorani-Arabic/1/dev"6    test: "datasets/Sorani-Arabic/1/test"7    level: "char"8    lowercase: False9    normalize: False10    max_sent_length: 5011    dataset_type: "plain"12 13    src:14        lang: "src"15        voc_limit: 10016        voc_min_freq: 517        level: "char"18    trg:19        lang: "trg"20        voc_limit: 10021        voc_min_freq: 522        level: "char"23    24training:25    random_seed: 4226    optimizer: "adam"27    learning_rate: 0.00128    learning_rate_min: 0.000229    weight_decay: 0.030    clip_grad_norm: 1.031    batch_size: 6432    scheduling: "plateau"33    patience: 534    decrease_factor: 0.535    early_stopping_metric: "loss"36    epochs: 2037    validation_freq: 100038    logging_freq: 10039    eval_metric: "bleu"40    model_dir: "models/Sorani-Arabic"41    overwrite: True42    shuffle: True43    use_cuda: True44    max_output_length: 5045    print_valid_sents: [0, 3, 6, 9]46    keep_best_ckpts: -147 48testing:49    n_best: 150    beam_size: 451    beam_alpha: 1.052    eval_metrics: ["bleu", "chrf", "sequence_accuracy"]53    max_output_length: 5054    batch_size: 1055    batch_type: "sentence"56    return_prob: "none"57 58model:59    initializer: "xavier_uniform" 60    init_gain: 1.0                61    bias_initializer: "zeros"     62    embed_initializer: "xavier_uniform" 63    embed_init_gain: 1.0          64    encoder:65        type: "transformer"66        num_layers: 667        num_heads: 868        embeddings:69            embedding_dim: 12870            scale: True71        # typically ff_size = 4 x hidden_size72        hidden_size: 12873        ff_size: 51274        dropout: 0.175        layer_norm: "pre"76    decoder:77        type: "transformer"78        num_layers: 679        num_heads: 880        embeddings:81            embedding_dim: 12882            scale: True83        # typically ff_size = 4 x hidden_size84        hidden_size: 12885        ff_size: 51286        dropout: 0.187        layer_norm: "pre"