The selected pre-train model is uncased_L-12_H-768_A-12

The running platform is Colab and you can run it on other Linux system.

The total number of data = 10000

The number distribution of Train: dev: test = 6:2:2

First experiment result eval_accuracy: 0.7702703

Second experiment result eval_accuracy: 0.7612613

Third experiment result eval_accuracy: 0.7742743

Average eval_accuracy by three times experiments: 0.76860196666

Range of change: (-0.00734066666, +0.00567233334)

Here is the code of running the run_classifier.py with detail training parameters which includes training and evaluation.

python run_classifier.py \
  --task_name=quora \
  --do_train=True \
  --do_eval=True \
  --do_predict=True \
  --data_dir=../Data/Model_train_dev_test_dataset/BERT_train_dev_test_dataset/ \
  --vocab_file=gs://cloud-tpu-checkpoints/bert/uncased_L-12_H-768_A-12/vocab.txt \
  --bert_config_file=gs://cloud-tpu-checkpoints/bert/uncased_L-12_H-768_A-12/bert_config.json \
  --init_checkpoint=gs://cloud-tpu-checkpoints/bert/uncased_L-12_H-768_A-12/bert_model.ckpt \
  --max_seq_length=128 \
  --train_batch_size=32 \
  --learning_rate=5e-5 \
  --num_train_epochs=2.0 \
  --output_dir=output 

Training script is referred from Sentence (and sentence-pair) classification tasks

Test script is referred from Prediction from classifier