def __init__(self, config):
super().__init__(config)
self.num_labels = config.num_labels
self.bert = AutoModel(config, add_pooling_layer=False) # BertModel
self.qa_outputs = nn.Linear(config.hidden_size, config.num_labels)
self.init_weights()
def __init__(self, config):
super(AutoModelForTokenClassification, self).__init__(config)
self.num_labels = config.num_labels
self.model = AutoModel(config)
self.dropout = DropoutMC(config.hidden_dropout_prob)
self.classifier = Linear(config.hidden_size, config.num_labels)
self.init_weights()
args = parser.parse_args()
tokenizer = AutoTokenizer.from_pretrained(args.model, use_fast=True)
model_config = config.ModelParameters(model_name=args.config_name,
hidden_size=args.embed_dim,
num_classes=3,
freeze_weights=False,
context_layers=(-1, ))
configuration = config.ParallelConfiguration(
model_parameters=model_config,
model=args.model,
sequence_max_len=args.seq_len,
save_path=args.output_dir,
batch_size=args.batch_size,
epochs=args.epochs,
device=torch.device(args.device),
tokenizer=tokenizer,
)
dataset = utils.load_file("../dataset/cached/wikipedia-topics")
dataloader = SmartParaphraseDataloader.build_batches(dataset,
16,
mode="sequence",
config=configuration)
model = AutoModel(args.model)
prune_rewire(args, sentence_model, dataloader, tokenizer)
