File size: 959 Bytes
0b21f29 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 |
# set train hyperparams
unset LD_PRELOAD
python3 run_mlm_flax.py \
--output_dir="./" \
--model_type="roberta" \
--config_name="./" \
--tokenizer_name="./" \
--dataset_filepath="/researchdisk/training_dataset_full" \
--save_tokenized_dataset_filepath="/researchdisk/training_dataset_full_tokenized_128" \
--max_seq_length="128" \
--pad_to_max_length \
--preprocessing_num_workers="96" \
--per_device_train_batch_size="64" \
--per_device_eval_batch_size="64" \
--adam_beta1="0.9" \
--adam_beta2="0.98" \
--adam_epsilon="1e-6" \
--learning_rate="2e-4" \
--warmup_steps="1500" \
--overwrite_output_dir \
--num_train_epochs="2" \
--save_strategy="steps" \
--save_steps="10000" \
--save_total_limit="5" \
--eval_steps="10000" \
--logging_steps="1000" \
--dtype="bfloat16" \
--push_to_hub \
--hub_model_id="Finnish-NLP/roberta-large-finnish-v2" \
--adafactor |