# en-ro/ro-en: 38.5/38.5 ->
CUDA_VISIBLE_DEVICES=0 python run_translation.py \
    --model_name_or_path facebook/mbart-large-en-ro  \
    --do_train \
    --do_eval \
    --dataset_name wmt16 \
    --dataset_config_name ro-en \
    --source_lang en_XX \
    --target_lang ro_RO \
    --output_dir ./baseline \
    --per_device_train_batch_size=4 \
    --per_device_eval_batch_size=4 \
    --overwrite_output_dir \
    --predict_with_generate \
    --apply-trp --trp-depths 1 --trp-p 0.1 --trp-lambdas 0.4 0.2 0.1