Add tokenizer

v3: multi-task with hard example mining

Training in progress, step 3105

Training in progress, step 3000

Training in progress, step 2800

Training in progress, step 2600

Training in progress, step 2400

Training in progress, step 2200

Training in progress, step 2000

Training in progress, step 1800

Training in progress, step 1600

Training in progress, step 1400

Training in progress, step 1200

Training in progress, step 1000

Training in progress, step 800

Training in progress, step 600

Training in progress, step 400

Training in progress, step 200

initial commit