Minor bug fix (#2443)

PaddleNLP/Research/MRQA2019-BASELINE/README.md

@@ -82,6 +82,7 @@ Where `parameters_to_restore` is the model parameters used in the evaluatation (
 | :------------- | :---------: | :----------: | :---------: | :----------: | :---------: | :----------: |:----------: |
 | baseline + EMA | 70.2/54.7 | 57.3/47.5 | 64.1/52.8 | 51.7/37.2 | 87.9/77.7 | 63.1/53.5 | 65.7 |
 | baseline woEMA | 69.9/54.6 | 57.0/47.3 | 64.0/52.8 | 51.8/37.4 | 87.8/77.6 | 63.0/53.4 | 65.6 |
+
 Note that we turn on exponential moving average (EMA) during training by default (in most cases EMA can improve performance) and save EMA parameters into the final checkpoint files. The predicted answers using EMA parameters are saved into `ema_predictions.json`.   
 
 

PaddleNLP/Research/MRQA2019-BASELINE/run_finetuning.sh

@@ -48,7 +48,7 @@ python -u src/run_mrqa.py --use_cuda true\
         --max_seq_len 512 \
         --bert_config_path ${BERT_BASE_PATH}/bert_config.json \
         --predict_file ${DATA_PATH_dev}/mrqa-combined.raw.json \
-        --do_lower_case false \
+        --do_lower_case true \
         --doc_stride 128 \
         --train_file ${DATA_PATH_train}/mrqa-combined.raw.json \
         --learning_rate 3e-5 \