Different Embeddding seqenence warning when I load model

[daizhonghao]

Environment

• OS [Linux]:
• Python Version: 3.6

Embedding sequence length is different when training and loading model

Hi,

First of all, I would like to thank you guys for your amazing job.
So, in my training model, I set BERTEmbedding sequence_length = 128,

embed = BERTEmbedding(BERT_PATH,task =kashgari.CLASSIFICATION,trainable=False,
sequence_length=128)
and my model is
model = BiLSTM_Model(embedding=embed,hyper_parameters=hyper)

And I save the model.

But when I load the model try to predict in another file predict,py, I also use this embedding, but I got 2 warnings:

W0116 10:31:27.545167 140598316177216 bert_embedding.py:126] seq_len: 128
W0116 10:31:27.771749 140598316177216 base_embedding.py:125] Sequence length will auto set at 95% of
in predict.py, this is how I load model:

embed = BERTEmbedding(BERT_PATH,task =kashgari.CLASSIFICATION,trainable=False,
sequence_length=128)
tokenizer = embed.tokenizer
raw_data = pd.read_csv("data.csv")
raw_data['cutted'] = raw_data['sentence'].apply(lambda x: tokenizer.tokenize(x))
model = load_model('bert_model',load_weights=False)
model.tf_model.load_weights('**.hdf5')

Why do I have the ### base_embedding warning when I load ### bert_embedding
How can I fix this problem? In my intuition, I should only receive a warning from BERT.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.