| python run_bart_dlm_flax.py \ | |
| --output_dir="./" \ | |
| --config_name="./" \ | |
| --tokenizer_name="./" \ | |
| --tokenized_dataset_filepath="/researchdisk/lm_training_dataset_tokenized" \ | |
| --preprocessing_num_workers="96" \ | |
| --max_seq_length="1024" \ | |
| --per_device_train_batch_size="4" \ | |
| --per_device_eval_batch_size="4" \ | |
| --learning_rate="4e-5" \ | |
| --weight_decay="0.01" \ | |
| --warmup_steps="10000" \ | |
| --overwrite_output_dir \ | |
| --num_train_epochs="5" \ | |
| --logging_steps="500" \ | |
| --save_steps="20000" \ | |
| --eval_steps="20000" \ | |
| --dtype="bfloat16" \ | |
| --use_auth_token \ | |
| --hub_model_id="Finnish-NLP/bart-large-finnish" \ | |
| --push_to_hub |