MLM_HF

sudo docker run -v /home/mamejose/Roberta:/workspace/Roberta --rm --gpus '"device=0,1,2,3,4,5,6,7"' -it mlm_transformers:1.0 python Roberta/src/run_mlm.py
--max_steps=1000000
--warmup_steps=20000 --dataloader_num_workers=4 --gradient_accumulation_steps=4 --save_steps=50000 --eval_steps=50000 --model_name_or_path=roberta-base --train_file /workspace/Roberta/data/02_splits/train.txt --validation_file=/workspace/Roberta/data/02_splits/dev.txt --per_device_train_batch_size 30
--per_device_eval_batch_size 60
--do_train --do_eval
--output_dir /workspace/Roberta/data/models/1mil_epochs/
--overwrite_output_dir