config.json

{
    "model_architecture" : {
        "token_embedding_dimension" : 128,
        "number_of_attention_heads" : 8,
        "number_of_layers" : 2,
        "ffn_units" : 512,
        "max_sentence_length" : 70,
        "dropout_rate" : 0.3
    },
    "training_settings" : {
        "batch_size" : 128,
        "epochs" : 70,
        "test_size" : 0.2,
        "warmup_steps" : 4000,
        "generate_training_stats" : true,
        "plot_training_stats" : false,
        "training_stats_save_dir" : "runs",
        "layernorm_epsilon_value" : 1e-6,
        "dropout_training" : true  
    },
    "data" : {
        "data_path" : "data//Translation//spa.txt",
        "approximate_source_tokens" : null,
        "approximate_target_tokens" : null
    },
    "save_paths" : {
        "model_save_path" : "model/english-to-spanish-ckpt",
        "source_tokenizers_save_path" : "model//english_tokenizer",
        "target_tokenizers_save_path" : "model//spanish_tokenizer"
    },
    "inference_settings" : {
        "visualise_attention_layer" : "decoder_layer_cross_attention_2",
        "attention_plot_save_dir" : "runs/exp"
    }
}