python run_qa.py \
    --model_name_or_path roberta-large \
    --dataset_name squad \
    --do_eval \
    --do_train \
    --evaluation_strategy steps \
    --eval_steps 500 \
    --learning_rate 3e-5 \
    --fp16 \
    --num_train_epochs 2 \
    --per_device_eval_batch_size 64 \
    --per_device_train_batch_size 16 \
    --max_seq_length 384 \
    --doc_stride 128 \
    --save_steps 1000 \
    --logging_steps 1 \
    --overwrite_output_dir \
    --run_name $RUNID \
    --output_dir $OUTDIR

Eval

bash

export CUDA_VISIBLE_DEVICES=0

MODEL=vuiseng9/roberta-l-squadv1.1
OUTDIR=eval-$(basename $MODEL)
WORKDIR=transformers/examples/pytorch/question-answering
cd $WORKDIR

nohup python run_qa.py  \
    --model_name_or_path  $MODEL \
    --dataset_name squad  \
    --do_eval  \
    --per_device_eval_batch_size 16  \
    --max_seq_length 384  \
    --doc_stride 128  \
    --overwrite_output_dir \
    --output_dir $OUTDIR 2>&1 | tee $OUTDIR/run.log &

bash

  eval_exact_match = 88.4674
  eval_f1          = 94.3001
  eval_samples     =   10790