mongolian-roberta-large / train_mlm.sh
bayartsogt's picture
train files
90cbb87
raw
history blame
538 Bytes
./run_mlm_flax.py \
--output_dir="${MODEL_DIR}" \
--model_type="roberta" \
--config_name="${MODEL_DIR}" \
--tokenizer_name="${MODEL_DIR}" \
--dataset_name="oscar" \
--dataset_config_name="unshuffled_deduplicated_mn" \
--max_seq_length="128" \
--weight_decay="0.01" \
--per_device_train_batch_size="64" \
--per_device_eval_batch_size="64" \
--learning_rate="3e-4" \
--warmup_steps="1000" \
--overwrite_output_dir \
--pad_to_max_length \
--num_train_epochs="300" \
--adam_beta1="0.9" \
--adam_beta2="0.98" \
--push_to_hub