BERT Sequence Classification

[andreaschandra]

• using BertWordPieceTokenizer to train title
• Pre-trained BERT from the default config

BertConfig {
  "attention_probs_dropout_prob": 0.1,
  "gradient_checkpointing": false,
  "hidden_act": "gelu",
  "hidden_dropout_prob": 0.1,
  "hidden_size": 768,
  "initializer_range": 0.02,
  "intermediate_size": 3072,
  "layer_norm_eps": 1e-12,
  "max_position_embeddings": 512,
  "model_type": "bert",
  "num_attention_heads": 12,
  "num_hidden_layers": 12,
  "pad_token_id": 0,
  "type_vocab_size": 2,
  "vocab_size": 30522
}

[andreaschandra]

@alamhanz

[andreaschandra]

## setup tokenizer and model
tokenizer = BertTokenizer.from_pretrained("bert-base-cased-finetuned-mrpc")
model = BertForSequenceClassification.from_pretrained("bert-base-cased-finetuned-mrpc")

[andreaschandra]

32GB memory usage

[andreaschandra]

pre-trained BERT failed, loss didn't decrease as expected