roberta_jan_128_ncc / run_128.sh
pere's picture
before training starts
61ffd98
raw
history blame
789 Bytes
python run_mlm_flax.py \
--output_dir="./" \
--model_type="roberta" \
--config_name="roberta-base" \
--tokenizer_name="NbAiLab/nb-roberta-base" \
--dataset_name="NbAiLab/NCC" \
--cache_dir="/mnt/disks/flaxdisk/cache/" \
--max_seq_length="128" \
--weight_decay="0.01" \
--per_device_train_batch_size="232" \
--per_device_eval_batch_size="232" \
--pad_to_max_length \
--learning_rate="6e-4" \
--warmup_steps="10000" \
--overwrite_output_dir \
--num_train_epochs="10" \
--adam_beta1="0.9" \
--adam_beta2="0.98" \
--adam_epsilon="1e-6" \
--logging_steps="1000" \
--save_steps="1000" \
--eval_steps="1000" \
--auth_token="True" \
--do_train \
--do_eval \
--dtype="bfloat16" \
--push_to_hub