fix run.sh

use proper tokenizer.json and change to batch_size=256

increaaase bs=128 and change dtype to fp32

remove old exp&model and lower steps to save

fix script

make it resumable training

update script

lower lr regarding explode gradient. also new seed number

switch back to batch_size = 32

change saving and eval steps

adjust batch_size and update run_mlm_flax script

adjust running script

reduce batch size

change to small batch size

change dtype to bfloat16

adjust script

update setting and batch size

change batch size and command

add runing script