Merge pull request huggingface#20 from stevezheng23/dev/zheng/quac

add quac codalab submission pipeline (cont.)

examples/run_quac.codalab.sh

@@ -24,16 +24,20 @@ alias python=python3
 git clone -b dev/zheng/quac --single-branch https://github.com/stevezheng23/transformers.git
 
 cd transformers
+cp examples/run_quac.py ./
+cp examples/util_quac.py ./
+cp examples/util_quac_evaluate.py ./
 
 mkdir data
 mkdir data/quac
 cp ../${INPUTFILE} data/quac/dev-v0.2.json
 
 mkdir output
 mkdir output/quac
-wget -P output/quac https://storage.googleapis.com/mrc_data/quac/roberta-large-quac-20191101.zip
-unzip output/quac/roberta-large-quac-20191101.zip -d output/quac/
-rm output/quac/roberta-large-quac-20191101.zip
+wget -P output/quac https://storage.googleapis.com/mrc_data/quac/roberta-large-quac-20191101.0.zip
+unzip output/quac/roberta-large-quac-20191101.0.zip -d output/quac/
+mv output/quac/roberta-large-quac-20191101.0 output/quac/roberta-large-quac
+rm output/quac/roberta-large-quac-20191101.0.zip
 
 CUDA_VISIBLE_DEVICES=0 python run_quac.py \
 --model_type=roberta \
@@ -48,7 +52,7 @@ CUDA_VISIBLE_DEVICES=0 python run_quac.py \
 --max_query_length=64 \
 --max_answer_length=30 \
 --doc_stride=128 \
---output_dir=output/quac/roberta-large-quac-20191101 \
+--output_dir=output/quac/roberta-large-quac \
 --per_gpu_eval_batch_size=12 \
 --per_gpu_train_batch_size=12 \
 --seed=1000 \
@@ -59,7 +63,7 @@ CUDA_VISIBLE_DEVICES=0 python run_quac.py \
 
 python utils_quac_evaluate.py \
 --val_file=data/quac/dev-v0.2.json \
---model_output=output/quac/roberta-large-quac-20191101/predictions_.json \
+--model_output=output/quac/roberta-large-quac/predictions_.json \
 --o=output/quac/pred.json
 
 cp output/quac/pred.json ../${OUTPUTFILE}