Currently I am working in NLP project involved with word embedding. I am using the pre-train Bert model from tensorflow hub, but when I feed my whole dataset into Bert layer