| ./run_mlm_flax.py \ | |
| --output_dir="./" \ | |
| --model_type="roberta" \ | |
| --config_name="./" \ | |
| --tokenizer_name="./" \ | |
| --train_file /mnt/disks/flaxdisk/corpus/train_1_4.json \ | |
| --validation_file /mnt/disks/flaxdisk/corpus/validation.json \ | |
| --cache_dir="/mnt/disks/flaxdisk/cache/" \ | |
| --max_seq_length="128" \ | |
| --weight_decay="0.01" \ | |
| --per_device_train_batch_size="200" \ | |
| --per_device_eval_batch_size="200" \ | |
| --learning_rate="6e-4" \ | |
| --warmup_steps="10000" \ | |
| --overwrite_output_dir \ | |
| --num_train_epochs="2" \ | |
| --adam_beta1="0.9" \ | |
| --adam_beta2="0.98" \ | |
| --logging_steps="10000" \ | |
| --save_steps="10000" \ | |
| --eval_steps="10000" \ | |
| --preprocessing_num_workers="64" \ | |
| --auth_token="True" \ | |
| --static_learning_rate="True" \ | |
| --dtype="bfloat16" \ | |
| --adafactor \ | |
| --push_to_hub | |