syntaxdot 0.5.0

Neural sequence labeler
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22

[input]
tokenizer = { bert = { vocab = "bert-base-german-cased-vocab.txt" } }

[biaffine]
labels = "sticker.biaffine_labels"
head = { dims = 50, head_bias = true, dependent_bias = false }
relation = { dims = 25, head_bias = true, dependent_bias = true }

[labeler]
labels = "sticker.labels"
encoders = [
  { name = "dep", encoder = { dependency = { encoder = { relativepos = "xpos" }, root_relation = "root" } } },
  { name = "lemma", encoder = { lemma = "form" } },
  { name = "pos", encoder = { sequence = "xpos" } },
]

[model]
parameters = "epoch-99"
pooler = "discard"
position_embeddings = "model"
pretrain_config = "bert_config.json"
pretrain_type = "bert"