Update Scripts to HF

SALT-NLP · Dec 15, 2022 · fbd80df · fbd80df
1 parent 1b71524
commit fbd80df
Show file tree

Hide file tree

Showing 14 changed files with 13 additions and 459 deletions.
diff --git a/coqa_exp/eval_bert.sh b/coqa_exp/eval_bert.sh
diff --git a/coqa_exp/eval_bert_adapters.sh b/coqa_exp/eval_bert_adapters.sh
diff --git a/coqa_exp/eval_combined_bert.sh b/coqa_exp/eval_combined_bert.sh
diff --git a/coqa_exp/eval_combined_roberta.sh b/coqa_exp/eval_combined_roberta.sh
diff --git a/coqa_exp/eval_in_bert.sh b/coqa_exp/eval_in_bert.sh
@@ -1,19 +1,18 @@
 export CUDA_VISIBLE_DEVICES=1
-#coqa CoQA_AAVE CoQA_AppE CoQA_ChcE CoQA_CollSgE CoQA_Multi
 
-for TRAIN_DIAL in coqa CoQA_AAVE CoQA_AppE CoQA_ChcE CoQA_CollSgE CoQA_IndE CoQA_Multi
+for TRAIN_DIAL in CoQA_SAE CoQA_AAVE CoQA_AppE CoQA_ChcE CoQA_CollSgE CoQA_IndE CoQA_Multi
 do
  for EVAL_DIAL in CoQA_SAE CoQA_AAVE CoQA_AppE CoQA_ChcE CoQA_CollSgE CoQA_IndE
  do
  python3 run_coqa.py --model_type bert \
- --model_name_or_path /data/wheld3/bert-$TRAIN_DIAL/ \
+ --model_name_or_path SALT-NLP/value-bert-$TRAIN_DIAL \
  --do_eval \
  --data_dir eval_data/$EVAL_DIAL/ \
  --train_file coqa-train-v1.0.json \
  --predict_file coqa-dev-v1.0.json \
  --learning_rate 3e-5 \
  --num_train_epochs 2 \
- --output_dir /data/wheld3/bert-$TRAIN_DIAL-$EVAL_DIAL/ \
+ --output_dir ./bert-$TRAIN_DIAL-$EVAL_DIAL/ \
  --do_lower_case \
  --per_gpu_train_batch_size 8 \
  --gradient_accumulation_steps 2 \

diff --git a/coqa_exp/eval_in_gold_bert.sh b/coqa_exp/eval_in_gold_bert.sh
@@ -1,23 +1,22 @@
 export CUDA_VISIBLE_DEVICES=0
-#coqa CoQA_AAVE CoQA_AppE CoQA_ChcE CoQA_CollSgE CoQA_Multi
 
 for TRAIN_DIAL in coqa CoQA_AAVE CoQA_ChcE CoQA_IndE CoQA_Multi
 do
  for EVAL_DIAL in CoQA_ChcE CoQA_IndE
  do
  python3 run_coqa.py --model_type bert \
- --model_name_or_path /data/wheld3/bert-$TRAIN_DIAL/ \
+ --model_name_or_path SALT-NLP/value-bert-$TRAIN_DIAL \
  --do_eval \
  --data_dir gold_data/$EVAL_DIAL/ \
  --train_file coqa-train-v1.0.json \
  --predict_file coqa-dev-v1.0.json \
  --learning_rate 3e-5 \
  --num_train_epochs 2 \
- --output_dir /data/wheld3/bert-$TRAIN_DIAL/ \
+ --output_dir ./bert-$TRAIN_DIAL/ \
  --do_lower_case \
  --per_gpu_train_batch_size 8 \
  --gradient_accumulation_steps 2 \
  --max_grad_norm -1 \
- --weight_decay 0.01 >> /data/wheld3/bert-$TRAIN_DIAL/in_gold_dial_$EVAL_DIAL
+ --weight_decay 0.01 >> ./bert-$TRAIN_DIAL/in_gold_dial_$EVAL_DIAL
  done
 done
diff --git a/coqa_exp/eval_in_gold_roberta.sh b/coqa_exp/eval_in_gold_roberta.sh
@@ -1,24 +1,22 @@
 export CUDA_VISIBLE_DEVICES=2
 
-#coqa CoQA_AAVE CoQA_AppE CoQA_ChcE CoQA_CollSgE CoQA_Multi
-
-for TRAIN_DIAL in coqa CoQA_AAVE CoQA_ChcE CoQA_IndE CoQA_Multi
+for TRAIN_DIAL in CoQA_SAE CoQA_AAVE CoQA_ChcE CoQA_IndE CoQA_Multi
 do
  for EVAL_DIAL in CoQA_ChcE CoQA_IndE
  do
  python3 run_coqa.py --model_type roberta \
- --model_name_or_path /data/wheld3/roberta-$TRAIN_DIAL/ \
+ --model_name_or_path SALT-NLP/value-roberta-$TRAIN_DIAL \
  --do_eval \
  --data_dir gold_data/$EVAL_DIAL/ \
  --train_file coqa-train-v1.0.json \
  --predict_file coqa-dev-v1.0.json \
  --learning_rate 3e-5 \
  --num_train_epochs 2 \
- --output_dir /data/wheld3/roberta-$TRAIN_DIAL/ \
+ --output_dir ./roberta-$TRAIN_DIAL/ \
  --do_lower_case \
  --per_gpu_train_batch_size 8 \
  --gradient_accumulation_steps 2 \
  --max_grad_norm -1 \
- --weight_decay 0.01 >> /data/wheld3/roberta-$TRAIN_DIAL/in_gold_dial_$EVAL_DIAL
+ --weight_decay 0.01 >> ./roberta-$TRAIN_DIAL/in_gold_dial_$EVAL_DIAL
  done
 done
diff --git a/coqa_exp/eval_in_roberta.sh b/coqa_exp/eval_in_roberta.sh
@@ -1,20 +1,18 @@
 export CUDA_VISIBLE_DEVICES=2
 
-#coqa CoQA_AAVE CoQA_AppE CoQA_ChcE CoQA_CollSgE CoQA_Multi
-
-for TRAIN_DIAL in coqa CoQA_AAVE CoQA_AppE CoQA_ChcE CoQA_CollSgE CoQA_IndE CoQA_Multi
+for TRAIN_DIAL in CoQA_SAE CoQA_AAVE CoQA_AppE CoQA_ChcE CoQA_CollSgE CoQA_IndE CoQA_Multi
 do
  for EVAL_DIAL in CoQA_SAE CoQA_AAVE CoQA_AppE CoQA_ChcE CoQA_CollSgE CoQA_IndE
  do
  python3 run_coqa.py --model_type roberta \
- --model_name_or_path /data/wheld3/roberta-$TRAIN_DIAL/ \
+ --model_name_or_path SALT-NLP/value-roberta-$TRAIN_DIAL \
  --do_eval \
  --data_dir eval_data/$EVAL_DIAL/ \
  --train_file coqa-train-v1.0.json \
  --predict_file coqa-dev-v1.0.json \
  --learning_rate 3e-5 \
  --num_train_epochs 2 \
- --output_dir /data/wheld3/roberta-$TRAIN_DIAL-$EVAL_DIAL/ \
+ --output_dir ./roberta-$TRAIN_DIAL-$EVAL_DIAL/ \
  --do_lower_case \
  --per_gpu_train_batch_size 8 \
  --gradient_accumulation_steps 2 \

diff --git a/coqa_exp/eval_roberta.sh b/coqa_exp/eval_roberta.sh