def __init__(self, cfg: DictConfig, trainer: Trainer = None):
self.setup_tokenizer(cfg.tokenizer)
super().__init__(cfg=cfg, trainer=trainer)
self.bert_model = get_lm_model(
pretrained_model_name=cfg.language_model.pretrained_model_name,
config_file=self.register_artifact('language_model.config_file',
cfg.language_model.config_file),
config_dict=OmegaConf.to_container(cfg.language_model.config)
if cfg.language_model.config else None,
checkpoint_file=cfg.language_model.lm_checkpoint,
vocab_file=self.register_artifact('tokenizer.vocab_file',
cfg.tokenizer.vocab_file)
if cfg.tokenizer is not None else None,
)
self.classifier = TokenClassifier(
hidden_size=self.bert_model.config.hidden_size,
num_classes=cfg.token_classifier.num_classes,
num_layers=cfg.token_classifier.num_layers,
activation=cfg.token_classifier.activation,
log_softmax=cfg.token_classifier.log_softmax,
dropout=cfg.token_classifier.dropout,
use_transformer_init=cfg.token_classifier.use_transformer_init,
)
self.loss = SpanningLoss()
def __init__(self, cfg: DictConfig, trainer: Trainer = None):
super().__init__(cfg=cfg, trainer=trainer)
self.classifier = TokenClassifier(
hidden_size=self.hidden_size,
num_classes=cfg.token_classifier.num_classes,
num_layers=cfg.token_classifier.num_layers,
activation=cfg.token_classifier.activation,
log_softmax=cfg.token_classifier.log_softmax,
dropout=cfg.token_classifier.dropout,
use_transformer_init=cfg.token_classifier.use_transformer_init,
)
self.loss = SpanningLoss()