ner_ontonotes_bert_emb.json

{
  "dataset_reader": {
    "class_name": "conll2003_reader",
    "data_path": "{DOWNLOADS_PATH}/ontonotes",
    "dataset_name": "ontonotes"
  },
  "dataset_iterator": {
    "class_name": "data_learning_iterator",
    "seed": 42
  },
  "chainer": {
    "in": ["x"],
    "in_y": ["y"],
    "pipe": [
      {
        "class_name": "transformers_bert_preprocessor",
        "vocab_file": "{BERT_PATH}/vocab.txt",
        "do_lower_case": false,
        "max_seq_length": 512,
        "in": ["x"],
        "out": ["x_tokens", "subword_tokens", "subword_tok_ids", "startofword_markers", "attention_mask"]
      },
      {
        "in": ["y"],
        "id": "tag_vocab",
        "class_name": "simple_vocab",
        "pad_with_zeros": true,
        "fit_on": ["y"],
        "save_path": "{MODEL_PATH}/tag.dict",
        "load_path": "{MODEL_PATH}/tag.dict",
        "out": ["y_ind"]
      },
      {
        "in": ["x_tokens"],
        "class_name": "mask",
        "out": ["mask"]
      },
      {
        "class_name": "transformers_bert_embedder",
        "id": "embedder",
        "bert_config_path": "{BERT_PATH}/bert_config.json",
        "truncate": false,
        "load_path": "{BERT_PATH}",
        "in": ["subword_tok_ids", "startofword_markers", "attention_mask"],
        "out": ["x_emb", "subword_emb", "max_emb", "mean_emb", "pooler_output"]
      },
      {
        "in": ["x_emb", "mask"],
        "in_y": ["y_ind"],
        "out": ["y_predicted"],
        "class_name": "ner",
        "main": true,
        "token_emb_dim": "#embedder.dim",
        "n_hidden_list": [256, 256, 256],
        "net_type": "rnn",
        "cell_type": "lstm",
        "use_cudnn_rnn": true,
        "n_tags": "#tag_vocab.len",
        "save_path": "{MODEL_PATH}/model",
        "load_path": "{MODEL_PATH}/model",
        "two_dense_on_top": true,
        "use_crf": true,
        "use_batch_norm": true,
        "embeddings_dropout": true,
        "top_dropout": true,
        "intra_layer_dropout": false,
        "l2_reg": 0,
        "learning_rate": 3e-3,
        "learning_rate_drop_patience": 3,
        "dropout_keep_prob": 0.7
      },
      {
        "ref": "tag_vocab",
        "in": ["y_predicted"],
        "out": ["tags"]
      }
    ],

    "out": ["x_tokens", "tags"]
  },
  "train": {
    "epochs": 100,
    "batch_size": 64,

    "metrics": [
      {
        "name": "ner_f1",
        "inputs": ["y", "tags"]
      },
      {
        "name": "ner_token_f1",
        "inputs": ["y", "tags"]
      }
    ],
    "validation_patience": 7,
    "val_every_n_epochs": 1,

    "log_every_n_batches": -1,
    "tensorboard_log_dir": "{MODEL_PATH}/logs",
    "show_examples": false,
    "class_name": "nn_trainer",
    "evaluation_targets": [
      "valid",
      "test"
    ]
  },
  "metadata": {
    "variables": {
      "ROOT_PATH": "~/.deeppavlov",
      "DOWNLOADS_PATH": "{ROOT_PATH}/downloads",
      "MODELS_PATH": "{ROOT_PATH}/models",
      "MODEL_PATH": "{MODELS_PATH}/ner_ontonotes_bert_emb",
      "BERT_PATH": "{DOWNLOADS_PATH}/bert_models/multi_cased_L-12_H-768_A-12_pt"
    },
    "download": [
      {
        "url": "http://files.deeppavlov.ai/deeppavlov_data/bert/multi_cased_L-12_H-768_A-12_pt.tar.gz",
        "subdir": "{DOWNLOADS_PATH}/bert_models"
      }
    ]
  }
}