dream

1
{
2
  "dataset_reader": {
3
    "class_name": "dp_sentseg_reader:SentSegDatasetReader",
4
    "data_path": "{DOWNLOADS_PATH}/sentseg_ru_subtitles_data",
5
    "dataset_name": "sentseg_ru",
6
    "provide_pos": false
7
  },
8
  "dataset_iterator": {
9
    "class_name": "data_learning_iterator"
10
  },
11
  "chainer": {
12
    "in": [
13
      "x"
14
    ],
15
    "in_y": [
16
      "y"
17
    ],
18
    "pipe": [
19
      {
20
        "class_name": "torch_transformers_ner_preprocessor",
21
        "vocab_file": "{TRANSFORMER}",
22
        "do_lower_case": true,
23
        "max_seq_length": 256,
24
        "max_subword_length": 15,
25
        "token_masking_prob": 0.0,
26
        "in": [
27
          "x"
28
        ],
29
        "out": [
30
          "x_tokens",
31
          "x_subword_tokens",
32
          "x_subword_tok_ids",
33
          "startofword_markers",
34
          "attention_mask"
35
        ]
36
      },
37
      {
38
        "id": "tag_vocab",
39
        "class_name": "simple_vocab",
40
        "unk_token": [
41
          "O"
42
        ],
43
        "pad_with_zeros": true,
44
        "save_path": "{MODEL_PATH}/tag.dict",
45
        "load_path": "{MODEL_PATH}/tag.dict",
46
        "fit_on": [
47
          "y"
48
        ],
49
        "in": [
50
          "y"
51
        ],
52
        "out": [
53
          "y_ind"
54
        ]
55
      },
56
      {
57
        "class_name": "torch_transformers_sequence_tagger",
58
        "n_tags": "#tag_vocab.len",
59
        "pretrained_bert": "{TRANSFORMER}",
60
        "attention_probs_keep_prob": 0.5,
61
        "return_probas": false,
62
        "encoder_layer_ids": [
63
          -1
64
        ],
65
        "optimizer": "AdamW",
66
        "optimizer_parameters": {
67
          "lr": 2e-05,
68
          "weight_decay": 1e-06,
69
          "betas": [
70
            0.9,
71
            0.999
72
          ],
73
          "eps": 1e-06
74
        },
75
        "clip_norm": 1.0,
76
        "min_learning_rate": 1e-07,
77
        "learning_rate_drop_patience": 30,
78
        "learning_rate_drop_div": 1.5,
79
        "load_before_drop": true,
80
        "save_path": "{MODEL_PATH}/model",
81
        "load_path": "{MODEL_PATH}/model",
82
        "in": [
83
          "x_subword_tok_ids",
84
          "attention_mask",
85
          "startofword_markers"
86
        ],
87
        "in_y": [
88
          "y_ind"
89
        ],
90
        "out": [
91
          "y_pred_ind"
92
        ]
93
      },
94
      {
95
        "ref": "tag_vocab",
96
        "in": [
97
          "y_pred_ind"
98
        ],
99
        "out": [
100
          "y_pred"
101
        ]
102
      }
103
    ],
104
    "out": [
105
      "x_tokens",
106
      "y_pred"
107
    ]
108
  },
109
  "train": {
110
    "epochs": 30,
111
    "batch_size": 64,
112
    "metrics": [
113
      {
114
        "name": "ner_f1",
115
        "inputs": [
116
          "y",
117
          "y_pred"
118
        ]
119
      },
120
      {
121
        "name": "ner_token_f1",
122
        "inputs": [
123
          "y",
124
          "y_pred"
125
        ]
126
      }
127
    ],
128
    "validation_patience": 100,
129
    "val_every_n_epochs": 1,
130
    "log_every_n_batches": 100,
131
    "show_examples": false,
132
    "pytest_max_batches": 2,
133
    "pytest_batch_size": 8,
134
    "evaluation_targets": [
135
      "valid"
136
    ],
137
    "class_name": "torch_trainer"
138
  },
139
  "metadata": {
140
    "variables": {
141
      "ROOT_PATH": "~/.deeppavlov",
142
      "DOWNLOADS_PATH": "{ROOT_PATH}/downloads",
143
      "MODELS_PATH": "{ROOT_PATH}/models",
144
      "TRANSFORMER": "DeepPavlov/rubert-base-cased-conversational",
145
      "MODEL_PATH": "{MODELS_PATH}/sentseg_ru_bert_torch_v0"
146
    },
147
    "download": [
148
      {
149
        "url": "http://files.deeppavlov.ai/deeppavlov_data/sentseg_ru_bert_torch_v0.tar.gz",
150
        "subdir": "{MODELS_PATH}"
151
      }
152
    ]
153
  }
154
}