PYTHONPATH=. python examples/pytorch/token_classification/finetune_token_classification.py \
    --task 'token-classification' \
    --trainer 'nlp-base-trainer' \
    --work_dir './tmp' \
    --model 'damo/nlp_structbert_backbone_base_std' \
    --train_dataset_name 'GeoGLUE' \
    --train_subset_name 'GeoETA' \
    --train_dataset_namespace 'damo' \
    --first_sequence 'tokens' \
    --eval_strategy by_step \
    --eval_interval 20 \
    --label 'ner_tags' \
    --sequence_length 128 \
    --preprocessor 'token-cls-tokenizer' \
    --preprocessor_padding 'max_length' \
    --max_epochs 2 \
    --mode 'inference' \
    --use_model_config True \
    --per_device_train_batch_size 32 \
    --train_data_worker 0 \
    --eval_data_worker 0 \
    --lr 3e-5 \