cjymz886 / gfner Goto Github PK

我发了一封邮件，希望获取到您所用的数据集~
发件人：[[email protected]]

Can you please share the dataset you used, the dataset I found so far is handled differently than yours and the code doesn't make sense!

Hi, I cannot reproduce the CoNLL2003 dataset results for f1=93.85 in your paper; the results only reached f1=91.

The parameters and model I used are from your code.:

vocab_size=30290
embedding_size=128
seq_len=127
random_seed = 2009
hidden_dropout_prob=0.1
learning_rate=3e-5
weight_decay=0.0
max_grad_norm=1.0
warmup=0.0
fix_bert_embeddings=False
min_num=1e-7
use_bert = True
batch_size=6
epochs=50
device = torch.device("cuda")

#bert_model = 'biobert-base-cased-v1.2'
#bert_model='biobert-large-cased-v1.1'
bert_model = 'bert-large-cased'

If available, could you provide me with a detailed list of the parameters you used for training?
Thank you !