transformers/examples/research_projects/seq2seq-distillation/finetune.sh

# the proper usage is documented in the README, you need to specify data_dir, output_dir and model_name_or_path
# run ./finetune.sh --help to see all the possible options
python finetune.py \
    --learning_rate=3e-5 \
    --fp16 \
    --gpus 1 \
    --do_train \
    --do_predict \
    --n_val 1000 \
    --val_check_interval 0.1 \
    "$@"