{
    "training_wkdir": "RESULTS_neuralTKF_6block_gpu",
    "rng_seednum": 9,

    "data_dir": "preprocess_data/DATA",
    "train_dset_splits": ["CHERRIES_split0"],
    "dev_dset_splits": ["CHERRIES_split1"],
    "test_dset_splits": ["CHERRIES_split1"],

    "toss_alignments_longer_than": 513,
    "batch_size": 8,
    "chunk_length": 512,

    "num_epochs": 30,
    "optimizer_config": {
        "init_value": 1e-4,
        "peak_value": 1e-4,
        "end_value": 1e-4,
        "warmup_steps": 0,
        "weight_decay": 1e-3,
        "every_k_schedule": 1
    },
    "early_stop_cond1_atol": 1e-3,
    "early_stop_cond2_gap": 0.1,
    "patience": 5,
    "update_grads": true,
    "use_scan_fns": false,

    "interms_for_tboard": {
        "sow_outputs": false,
        "embeddings": false,
        "gradients": false,
        "weights": false,
        "optimizer": false,
        "forward_pass_outputs": false,
        "attn_weights": false
    },
    "save_arrs": false,
    "save_per_sample_losses": false,
    "record_metrics_every_n_steps": 500,
    "checkpoint_every_t_seconds": 1800,

    "anc_model_type": "transformer",
    "anc_enc_config": {"initial_embed_module": "EmbeddingWithPadding",
                       "first_block_module": "RoPETransfBlock",
                       "subsequent_block_module": "RoPETransfBlock",
                       "hidden_dim": 756,
                       "num_blocks": 6,
                       "num_heads": 6,
                       "dropout": 0.2,
                       "max_position_embeddings": 3000
                       },

    "desc_model_type": "transformer",
    "desc_dec_config": {"initial_embed_module": "EmbeddingWithPadding",
                       "first_block_module": "RoPETransfBlock",
                       "subsequent_block_module":"RoPETransfBlock",
                       "hidden_dim": 756,
                       "num_blocks": 6,
                       "num_heads": 6,
                       "dropout": 0.2,
                       "max_position_embeddings": 3000
                       },

    "pred_model_type": "neural_hmm",
    "pred_config": {
        "load_all": false,
        "subst_model_type": "f81",
        "indel_model_type": "tkf92",

        "global_or_local": {"equl_dist": "local",
                            "rate_mult": "local",
                            "tkf_rates": "local",
                            "tkf92_frag_size": "local"},

        "emissions_postproc_model_type": "feedforward",
        "emissions_postproc_config": {"use_anc_emb": true,
                                      "use_desc_emb": true,
                                      "use_prev_align_info": false,
                                      "layer_sizes": [500],
                                      "dropout": 0.20,
                                      "normalize_seq_embeddings_before_block": true},

        "transitions_postproc_model_type": "feedforward",
        "transitions_postproc_config": {"use_anc_emb": true,
                                        "use_desc_emb": true,
                                        "use_prev_align_info": true,
                                        "layer_sizes": [500],
                                        "dropout": 0.20,
                                        "normalize_seq_embeddings_before_block": true},

        "regularization_rates": {"equl_dist":0,
                                 "rate_mult":0,
                                 "tkf_lambda":0,
                                 "tkf_mu":0,
                                 "tkf92_frag_size":0},

        "times_from": "t_per_sample"

    }

}