dream

1
{
2
  "dataset_reader": {
3
    "class_name": "sq_reader",
4
    "data_path": "{DOWNLOADS_PATH}/wikipedia_entity_detection/wiki_ent_eng.pickle"
5
  },
6
  "dataset_iterator": {
7
    "class_name": "data_learning_iterator"
8
  },
9
  "chainer": {
10
    "in": ["x"],
11
    "in_y": ["y"],
12
    "pipe": [
13
      {
14
        "class_name": "torch_transformers_ner_preprocessor",
15
        "vocab_file": "{TRANSFORMER}",
16
        "do_lower_case": true,
17
        "max_seq_length": 512,
18
        "max_subword_length": 15,
19
        "token_masking_prob": 0.0,
20
        "in": ["x"],
21
        "out": ["x_tokens", "x_subword_tokens", "x_subword_tok_ids", "startofword_markers", "attention_mask", "tokens_offsets"]
22
      },
23
      {
24
        "id": "tag_vocab",
25
        "class_name": "simple_vocab",
26
        "unk_token": ["O"],
27
        "pad_with_zeros": true,
28
        "save_path": "{MODEL_PATH}/tag.dict",
29
        "load_path": "{MODEL_PATH}/tag.dict",
30
        "fit_on": ["y"],
31
        "in": ["y"],
32
        "out": ["y_ind"]
33
      },
34
      {
35
        "class_name": "torch_transformers_sequence_tagger",
36
        "n_tags": "#tag_vocab.len",
37
        "pretrained_bert": "{TRANSFORMER}",
38
        "attention_probs_keep_prob": 0.5,
39
        "encoder_layer_ids": [-1],
40
        "optimizer": "AdamW",
41
        "optimizer_parameters": {
42
          "lr": 2e-05,
43
          "weight_decay": 1e-06,
44
          "betas": [0.9, 0.999],
45
          "eps": 1e-06
46
        },
47
        "clip_norm": 1.0,
48
        "min_learning_rate": 1e-07,
49
        "learning_rate_drop_patience": 4,
50
        "learning_rate_drop_div": 1.5,
51
        "load_before_drop": true,
52
        "save_path": "{MODEL_PATH}/model",
53
        "load_path": "{MODEL_PATH}/model",
54
        "in": ["x_subword_tok_ids", "attention_mask", "startofword_markers"],
55
        "in_y": ["y_ind"],
56
        "out": ["y_pred_ind", "probas"]
57
      },
58
      {
59
        "ref": "tag_vocab",
60
        "in": ["y_pred_ind"],
61
        "out": ["y_pred"]
62
      }
63
    ],
64
    "out": ["x_tokens", "tokens_offsets", "y_pred", "probas"]
65
  },
66
  "train": {
67
    "epochs": 5,
68
    "batch_size": 30,
69
    "valid_batch_size": 30,
70
    "metrics": [
71
      {
72
        "name": "ner_f1",
73
        "inputs": ["y", "y_pred"]
74
      },
75
      {
76
        "name": "ner_token_f1",
77
        "inputs": ["y", "y_pred"]
78
      }
79
    ],
80
    "validation_patience": 10,
81
    "val_every_n_batches": 100,
82
    "log_every_n_batches": 100,
83
    "evaluation_targets": ["valid", "test"],
84
    "class_name": "torch_trainer"
85
  },
86
  "metadata": {
87
    "variables": {
88
      "ROOT_PATH": "~/.deeppavlov",
89
      "DOWNLOADS_PATH": "{ROOT_PATH}/downloads",
90
      "TRANSFORMER": "bert-base-uncased",
91
      "MODEL_PATH": "{ROOT_PATH}/models/dialog_entity_detection"
92
    },
93
    "requirements": [
94
      "{DEEPPAVLOV_PATH}/requirements/transformers.txt",
95
      "{DEEPPAVLOV_PATH}/requirements/pytorch.txt",
96
      "{DEEPPAVLOV_PATH}/requirements/torchcrf.txt"
97
    ],
98
    "download": [
99
      {
100
        "url": "http://files.deeppavlov.ai/deeppavlov_data/dialog_entity_detection/dialog_entity_detection_dream.tar.gz",
101
        "subdir": "{MODEL_PATH}"
102
      }
103
    ]
104
  }
105
}
106