{
    "dataset_reader": {
        "type": "multitask",
        "readers": {
            "ner": {
                "type": "wikiann",
                "token_indexers": {
                    "transformer": {
                        "type": "pretrained_transformer_mismatched",
                        "max_length": 512,
                        "model_name": "MLRS/BERTu"
                    }
                }
            }
        }
    },
    "model": {
        "type": "multitask",
        "arg_name_mapping": {
            "backbone": {
                "tokens": "text",
                "words": "text"
            }
        },
        "backbone": {
            "type": "embedder_and_mask",
            "text_field_embedder": {
                "token_embedders": {
                    "transformer": {
                        "type": "pretrained_transformer_mismatched_with_dropout",
                        "last_layer_only": false,
                        "layer_dropout": 0.1,
                        "max_length": 512,
                        "model_name": "MLRS/BERTu",
                        "tokenizer_kwargs": {},
                        "train_parameters": true
                    }
                }
            }
        },
        "heads": {
            "ner": {
                "type": "crf_tagger",
                "calculate_span_f1": true,
                "constrain_crf_decoding": true,
                "dropout": 0.2,
                "encoder": {
                    "type": "pass_through",
                    "input_dim": 768
                },
                "include_start_end_transitions": false,
                "label_encoding": "BIO"
            }
        }
    },
    "train_data_path": {
        "ner": "panx_dataset/mt/train"
    },
    "validation_data_path": {
        "ner": "panx_dataset/mt/dev"
    },
    "trainer": {
        "callbacks": [
            {
                "tensorboard_writer": {
                    "should_log_learning_rate": true,
                    "should_log_parameter_statistics": true
                },
                "type": "tensorboard"
            }
        ],
        "cuda_device": 0,
        "grad_norm": 5,
        "learning_rate_scheduler": {
            "type": "ulmfit_sqrt",
            "affected_group_count": 2,
            "decay_factor": 0.05,
            "discriminative_fine_tuning": true,
            "factor": 5,
            "gradual_unfreezing": true,
            "model_size": 1,
            "start_step": 2,
            "warmup_steps": 2
        },
        "num_epochs": 200,
        "optimizer": {
            "type": "huggingface_adamw",
            "betas": [
                0.9,
                0.999
            ],
            "correct_bias": false,
            "lr": 0.0005,
            "parameter_groups": [
                [
                    [
                        "text_field_embedder.*transformer_model.embeddings.*_embeddings.*",
                        "text_field_embedder.*transformer_model.encoder.*.(key|query|value|dense).weight"
                    ],
                    {}
                ],
                [
                    [
                        "text_field_embedder.*transformer_model.embeddings.LayerNorm.*",
                        "text_field_embedder.*transformer_model.encoder.*.output.LayerNorm.*",
                        "text_field_embedder.*transformer_model.encoder.*.(key|query|value|dense).bias",
                        "text_field_embedder.*transformer_model.pooler.dense.bias"
                    ],
                    {
                        "weight_decay": 0
                    }
                ],
                [
                    [
                        "text_field_embedder.*._scalar_mix.*",
                        "text_field_embedder.*transformer_model.pooler.dense.weight",
                        "_head_sentinel",
                        "head_arc_feedforward._linear_layers.*.weight",
                        "child_arc_feedforward._linear_layers.*.weight",
                        "head_tag_feedforward._linear_layers.*.weight",
                        "child_tag_feedforward._linear_layers.*.weight",
                        "arc_attention._weight_matrix",
                        "tag_bilinear.weight",
                        "tag_projection_layer._module.weight",
                        "crf",
                        "linear.weight",
                        "tagger_linear.weight"
                    ],
                    {}
                ],
                [
                    [
                        "head_arc_feedforward._linear_layers.*.bias",
                        "child_arc_feedforward._linear_layers.*.bias",
                        "head_tag_feedforward._linear_layers.*.bias",
                        "child_tag_feedforward._linear_layers.*.bias",
                        "arc_attention._bias",
                        "tag_bilinear.bias",
                        "tag_projection_layer._module.bias",
                        "linear.bias",
                        "tagger_linear.bias"
                    ],
                    {
                        "weight_decay": 0
                    }
                ]
            ],
            "weight_decay": 0.01
        },
        "patience": 20,
        "validation_metric": [
            "+ner_f1-measure-overall"
        ]
    },
    "data_loader": {
        "type": "multitask",
        "scheduler": {
            "type": "unbalanced_homogeneous_roundrobin",
            "batch_size": 64,
            "dataset_sizes": {
                "ner": 100
            }
        },
        "shuffle": true
    },
    "numpy_seed": 1337,
    "pytorch_seed": 133,
    "random_seed": 13370,
    "validation_data_loader": {
        "type": "multitask",
        "scheduler": {
            "type": "homogeneous_roundrobin",
            "batch_size": 64
        },
        "shuffle": true
    }
}