script_running_notes.txt

server 2
2977136  BBC/Cosmos/RACE/logiqa-roberta-large-baseline/roberta-large-mnli/roberta-reclor-logical-equivalence-lreasoner/our-model-roberta-large-pararule-depth-1/our-model-roberta-large-v5-mnli/our-model-pararule-plus-logic-equivalence-depth-1-v4
2979905  ReClor/LogiQA/synthetic/logiqa-roberta-large-our-amr-method/our-model-roberta-large-mnli/RoBERTa-large-lreasoner-sst2/lreasoner-roberta-large-pararule-depth-1/our-model-roberta-large-v3-mrpc/our-model-roberta-large-v4-rte/our-model-roberta-large-v4-qnli-bs-32/our-model-rte-v5/our-model-qqp-v5/lreasoner-pararule-plus-depth-1
1017037  Cosmos/roberta-large-stsb/RoBERTa-large-mrpc/RoBERTa-large-qqp/RoBERTa-large-lreasoner-qqp/roberta-large-qnli-bs-8/roberta-large-pararule-bs-16/our-model-roberta-large-v2-rte/our-model-reclor-v4/our-model-logiqa-v5/lreasoner-reclor-bs-1/our-model-pararule-plus-depth-1-v4
1017825  Wikihop/roberta-large-synthetic-pretraining/roberta-logical-equivalence-reclor/roberta-reclor/our-model-RoBERTa-large-mrpc/synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp/RoBERTa-large-lreasoner-qnli/roberta-large-qnli-bs-16-max-length-128/roberta-large-pararule-bs-16-lr-5e-5/our-model-roberta-large-v4-qnli/our-model-qnli-v5/lreasoner-pararule-plus-logical-equivalence-depth-1
220314   RoBERTa-large-rte/synthetic-logical-equivalence-finetuned-RoBERTa-large-rte/RoBERTa-large-wnli/RoBERTa-large-qnli/RoBERTa-large-Lreasoner-wnli/roberta-large-rte-bs-16/roberta-large-qnli-bs-16/roberta-large-pararule-bs-8-lr-5e-5/roberta-large-qnli-bs-16-max-length-128/roberta-large-pararule-bs-16-lr-5e-5/our-model-roberta-large-v4-qqp/roberta-large-pararule-plus-depth-1
223070   synthetic-logical-equivalence-finetuned-RoBERTa-large-wnli/synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli/RoBERTa-large-Lreasoner-RTE/roberta-large-lreasoner-logical-equivalence-logicqa/roberta-large-pararule-depth-1/our-model-qqp-v4/roberta-large-pararule-plus-depth-1-logical-equivalence
682487   RoBERTa-large-cola/RoBERTa-large-SST2/RoBERTa-large-LReasoner-MNLI/our-method-roberta-large-rte-bs-16/our-method-roberta-large-qnli-bs-16/
685058   synthetic-logical-equivalence-finetuned-RoBERTa-large-cola/synthetic-logical-equivalence-finetuned-RoBERTa-large-sst2/RoBERTa-large-LReasoner-MRPC/lreasoner-roberta-large-qnli-bs-16/roberta-large-qnli-bs-32/our-model-reclor-v5-bs-1/

HGX1
contraposition
 roberta-large-contraposition-mnli/
 roberta-large-contraposition-mrpc/roberta-large-contraposition-reclor/
 roberta-large-contraposition-rte/roberta-large-contraposition-logiqa/
 roberta-large-contraposition-qnli/roberta-large-contraposition-qqp/
4021880 roberta-large-contraposition-qqp-bs-32/16

contraposition+double negation
 roberta-large-contraposition-double-negation-mnli/
4046614 roberta-large-contraposition-double-negation-mnli-bs-16/8
 roberta-large-contraposition-double-negation-mrpc/roberta-large-contraposition-double-negation-qqp/
 roberta-large-contraposition-double-negation-rte/roberta-large-contraposition-double-negation-logiqa/
 roberta-large-contraposition-double-negation-qnli/
 roberta-large-contraposition-double-negation-reclor/

contraposition+double negation+commutative
 roberta-large-contraposition-double-negation-commutative-mnli/
 roberta-large-contraposition-double-negation-commutative-mrpc/
 roberta-large-contraposition-double-negation-commutative-rte/
 roberta-large-contraposition-double-negation-commutative-qnli/
 roberta-large-contraposition-double-negation-commutative-reclor/
 roberta-large-contraposition-double-negation-commutative-logiqa/

contraposition+double negation+implication
 roberta-large-contraposition-double-negation-implication-mnli-filled/
 roberta-large-contraposition-double-negation-implication-mrpc-filled/roberta-large-contraposition-double-negation-implication-logiqa-filled/
 roberta-large-contraposition-double-negation-implication-rte-filled/
 roberta-large-contraposition-double-negation-implication-qnli-filled/
4042248 roberta-large-contraposition-double-negation-implication-qqp-filled-bs-32/16
 roberta-large-contraposition-double-negation-implication-reclor-filled/

contraposition+double negation+implication+commutative
4039412 v4-first-stage-finetuning/roberta-large-contraposition-double-negation-implication-commutative-reclor
4046988 roberta-large-contraposition-double-negation-implication-commutative-qnli


negative sampling
 our model with positive:negative = 1:2 first stage finetuning
4043958 1-2-reclor
4040269 1-2-logiqa
4038342 1-2-mnli
 1-2-mrpc/
 1-2-rte/
4019763 1-2-qnli
4018413 1-2-qqp

 our model with positive:negative = 1:3 first stage finetuning
4044901 1-3-reclor
4036928 1-3-logiqa
4016700 1-3-mrpc/
4020794 1-3-rte/
88627 1-3-mnli
89938 1-3-qnli
90803 1-3-qqp

alberta
4045769 alberta-reclor

deberta
4043034 deberta-reclor

g.triples.remove((z5, ':polarity', '-'))

z0 = swap_condition[0]


### RoBERTa-synthetic-logical-equivalence-pretraining-our-model
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../Synthetic_xfm_t5wtense_logical_equivalence_train.csv \
  --validation_file ../Synthetic_xfm_t5wtense_logical_equivalence_validation.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/roberta-large/


### RoBERTa-synthetic-logical-equivalence-pretraining-our-model-version-2
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_our_model_logical_equivalence_train_version2.csv \
  --validation_file ../output_result/Synthetic_our_model_logical_equivalence_validation_version2.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 30 \
  --output_dir Transformers/roberta-large-our-model-v2/


### RoBERTa-synthetic-logical-equivalence-pretraining-our-model-version-3
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_train_v3.csv \
  --validation_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_validation_v3.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 30 \
  --output_dir Transformers/roberta-large-our-model-v3/


### RoBERTa-synthetic-logical-equivalence-pretraining-contraposition-law
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_train_contraposition.csv \
  --validation_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_validation_contraposition.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/roberta-large-contraposition/

### RoBERTa-synthetic-logical-equivalence-pretraining-contraposition-double-negation-law
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_xfm_t5wtense_le_contra_double_neg_list_train.csv \
  --validation_file ../output_result/Synthetic_xfm_t5wtense_le_contra_double_neg_list_dev.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/roberta-large-contraposition-double-negation/

### RoBERTa-synthetic-logical-equivalence-pretraining-contraposition-double-negation-commutative-law
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_xfm_t5wtense_le_contra_double_neg_commuta_list_train.csv \
  --validation_file ../output_result/Synthetic_xfm_t5wtense_le_contra_double_neg_commuta_list_dev.csv \
  --max_length 256 \
  --per_device_train_batch_size 8 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/roberta-large-contraposition-double-negation-commutative/

  (batch size=16/32 failed)


### RoBERTa-synthetic-logical-equivalence-pretraining-contraposition-double-negation-implication-law
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_xfm_t5wtense_le_contra_double_neg_commuta_impli_list_train.csv \
  --validation_file ../output_result/Synthetic_xfm_t5wtense_le_contra_double_neg_commuta_impli_list_dev.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/roberta-large-contraposition-double-negation-implication/

### RoBERTa-synthetic-logical-equivalence-pretraining-contraposition-double-negation-implication-law
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_xfm_t5wtense_le_contra_double_neg_commuta_impli_list_train_filled.csv \
  --validation_file ../output_result/Synthetic_xfm_t5wtense_le_contra_double_neg_commuta_impli_list_dev_filled.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/roberta-large-contraposition-double-negation-implication-filled/

### RoBERTa-synthetic-logical-equivalence-pretraining-our-model-version-4
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_train_v4.csv \
  --validation_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_validation_v4.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/roberta-large-our-model-v4/

### RoBERTa-synthetic-logical-equivalence-pretraining-our-model-version-4-positive-negative-1-2
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_train_v4_negative_samples_1_2.csv \
  --validation_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_validation_v4.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/roberta-large-our-model-v4-pos-neg-1-2/

### RoBERTa-synthetic-logical-equivalence-pretraining-our-model-version-4-positive-negative-1-3
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_train_v4_negative_samples_1_3.csv \
  --validation_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_validation_v4.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/roberta-large-our-model-v4-pos-neg-1-3/


### RoBERTa-large-Max-Tegmark
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../Max_Tegmark/equivalence-full-train.csv \
  --validation_file ../Max_Tegmark/equivalence-full-dev.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/Max_Tegmark/roberta-large-Max-Tegmark/

### RoBERTa-large-our-model-v4-Max-Tegmark
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path Transformers/roberta-large-our-model-v4/ \
  --train_file ../Max_Tegmark/equivalence-full-train.csv \
  --validation_file ../Max_Tegmark/equivalence-full-dev.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/Max_Tegmark/roberta-large-our-model-v4-Max-Tegmark/

### RoBERTa-large-our-model-v5-Max-Tegmark
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path Transformers/roberta-large-our-model-v5/ \
  --train_file ../Max_Tegmark/equivalence-full-train.csv \
  --validation_file ../Max_Tegmark/equivalence-full-dev.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/Max_Tegmark/roberta-large-our-model-v5-Max-Tegmark/

### RoBERTa-large-lreasoner-Max-Tegmark
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path Transformers/roberta-large-lreasoner/ \
  --train_file ../Max_Tegmark/equivalence-full-train.csv \
  --validation_file ../Max_Tegmark/equivalence-full-dev.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/Max_Tegmark/roberta-large-lreasoner-Max-Tegmark/

### RoBERTa-synthetic-logical-equivalence-pretraining-our-model-version-5
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_train_v5.csv \
  --validation_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_validation_v5.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 30 \
  --output_dir Transformers/roberta-large-our-model-v5/

### RoBERTa-synthetic-logical-equivalence-pretraining-lreasoner
python run_glue_no_trainer.py \
  --seed 2021 \
  --model_name_or_path roberta-large \
  --train_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_train.csv \
  --validation_file ../output_result/Synthetic_xfm_t5wtense_logical_equivalence_validation.csv \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/roberta-large-lreasoner/
##################################################################################################

### RoBERTa-large-MNLI
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path roberta-large \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/roberta-large/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-logical-equivalence-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v2/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-logical-equivalence-finetuned-roberta-large-v2/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI-v3
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v3/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-logical-equivalence-finetuned-roberta-large-v3/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI-comtraposition-law
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-le-contraposition-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI-comtraposition-double-negation-law
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-le-contra-double-neg-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI-comtraposition-double-negation-law
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 8 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-le-contra-double-neg-finetuned-roberta-large-bs-8/
(failed on bs=16/32)

### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI-comtraposition-double-negation-law-commutative
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation-commutative/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-le-contra-double-neg-commu-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI-comtraposition-double-negation-law-implication-fileld
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation-implication-filled/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-le-contra-double-neg-impli-finetuned-roberta-large-filled/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI-v4
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-logical-equivalence-finetuned-roberta-large-v4/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI-v4-positive-negative-1-2
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4-pos-neg-1-2/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-logical-equivalence-finetuned-roberta-large-v4-pn-1-2/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI-v4-positive-negative-1-3
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4-pos-neg-1-3/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-logical-equivalence-finetuned-roberta-large-v4-pn-1-3/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI-v5
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v5/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/synthetic-logical-equivalence-finetuned-roberta-large-v5/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-MNLI-lreasoner
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-lreasoner/ \
  --task_name mnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/MNLI/lreasoner-synthetic-logical-equivalence-finetuned-roberta-large/

##################################################################################################

### RoBERTa-large-mrpc
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path roberta-large \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/roberta-large/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/synthetic-logical-equivalence-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v2/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/synthetic-logical-equivalence-finetuned-roberta-large-v2/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v3/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/synthetic-logical-equivalence-finetuned-roberta-large-v3/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/synthetic-le-contraposition-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc-contraposition-double-negation
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/synthetic-le-contra-double-neg-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc-contraposition-double-negation-commutative
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation-commutative/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/synthetic-le-contra-double-neg-commu-finetuned-roberta-large/
(failed on bs=32)

### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc-contraposition-double-negation-commutative
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation-implication-filled/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/synthetic-le-contra-double-neg-impli-finetuned-roberta-large-filled/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/synthetic-logical-equivalence-finetuned-roberta-large-v4/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4-pos-neg-1-2/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/roberta-large-le-our-model-mrpc-v4-pn-1-2/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc-positive-negative
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4-pos-neg-1-3/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/roberta-large-le-our-model-mrpc-v4-pn-1-3/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v5/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/synthetic-logical-equivalence-finetuned-roberta-large-v5/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-mrpc-lreasoner
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-lreasoner/ \
  --task_name mrpc \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/mrpc/lreasoner-synthetic-logical-equivalence-finetuned-roberta-large/

##################################################################################################

### RoBERTa-large-rte
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path roberta-large \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/roberta-large-bs-16/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/synthetic-logical-equivalence-finetuned-roberta-large-bs-16/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v2/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/synthetic-logical-equivalence-finetuned-roberta-large-bs-16-v2/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/synthetic-le-contraposition-finetuned-roberta-large-bs-16/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte-contra-double-neg
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/synthetic-le-contra-double-neg-finetuned-roberta-large-bs-16/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte-contra-double-neg-commutative
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation-commutative/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 30 \
  --output_dir Transformers/rte/synthetic-le-contra-double-neg-commu-finetuned-roberta-large-bs-32/

(failed on bs=16,8,4,1,2)
32

### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte-contra-double-neg-implication
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation-implication-filled/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/synthetic-le-contra-double-neg-commu-finetuned-roberta-large-bs-16-filled/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/synthetic-logical-equivalence-finetuned-roberta-large-bs-16-v4/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v5/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/synthetic-logical-equivalence-finetuned-roberta-large-bs-16-v5/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte-lreasoner
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-lreasoner/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/lreasoner-synthetic-logical-equivalence-finetuned-roberta-large-bs-16/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/synthetic-logical-equivalence-finetuned-roberta-large/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte-v4
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/synthetic-logical-equivalence-finetuned-roberta-large-bs-32-v4/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte-v4-positive-negative-1-2
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4-pos-neg-1-2/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/roberta-large-le-our-model-rte-v4-pn-1-2/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte-v4-positive-negative-1-3
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4-pos-neg-1-3/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/roberta-large-le-our-model-rte-v4-pn-1-3/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte-v5
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v5/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/synthetic-logical-equivalence-finetuned-roberta-large-bs-32-v5/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-rte-lreasoner
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-lreasoner/ \
  --task_name rte \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/rte/lreasoner-synthetic-logical-equivalence-finetuned-roberta-large/

##################################################################################################

## RoBERTa-large-qnli-trainer
python run_glue.py \
  --model_name_or_path roberta-large \
  --task_name qnli \
  --do_train \
  --do_eval \
  --max_seq_length 128 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/roberta-large-trainer-bs-32/

### RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path roberta-large \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 8 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/roberta-large-bs-8/


### RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path roberta-large \
  --task_name qnli \
  --max_length 128 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/roberta-large-bs-16-max_length-128/

### RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path roberta-large \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 5e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/roberta-large-bs-16-lr-5e-5/

python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path roberta-large \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/roberta-large-bs-16/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-logical-equivalence-finetuned-roberta-large-bs-16/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v2/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-logical-equivalence-finetuned-roberta-large-bs-16-v2/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v3/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-logical-equivalence-finetuned-roberta-large-bs-16-v3/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-le-contraposition-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli-contra-double-neg
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-le-contra-double-neg-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli-contra-double-neg-commu
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation-commutative/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-le-contra-double-neg-commu-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli-contra-double-neg-implication-filled
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation-implication-filled/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-le-contra-double-neg-impli-finetuned-roberta-large-filled/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-logical-equivalence-finetuned-roberta-large-bs-16-v4/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli-a100
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-logical-equivalence-finetuned-roberta-large-bs-16-v4/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli-positive-negative-1-2
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4-pos-neg-1-2/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/roberta-large-le-our-model-logiqa-v4-pn-1-2/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli-positive-negative-1-3
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4-pos-neg-1-3/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/roberta-large-le-our-model-logiqa-v4-pn-1-3/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli-lreasoner
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-lreasoner/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/lreasoner-synthetic-logical-equivalence-finetuned-roberta-large-bs-16/

### RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path roberta-large \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/roberta-large/



### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-logical-equivalence-finetuned-roberta-large/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-logical-equivalence-finetuned-roberta-large-bs-32-v4/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v5/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/synthetic-logical-equivalence-finetuned-roberta-large-bs-32-v5/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qnli-lreasoner
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-lreasoner/ \
  --task_name qnli \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qnli/lreasoner-synthetic-logical-equivalence-finetuned-roberta-large/

##################################################################################################

### RoBERTa-large-qqp
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path roberta-large \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/roberta-large/



### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/synthetic-logical-equivalence-finetuned-roberta-large/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v2/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/synthetic-logical-equivalence-finetuned-roberta-large-v2/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v3/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/synthetic-logical-equivalence-finetuned-roberta-large-v3/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/synthetic-logical-equivalence-finetuned-roberta-large-v4/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/synthetic-le-contraposition-finetuned-roberta-large-bs-16/

(failed on bs=32)

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp-contraposition-double-negation
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/synthetic-le-contra-double-neg-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp-contraposition-double-negation-commutative
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation-commutative/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/synthetic-le-contra-double-neg-commu-finetuned-roberta-large/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp-contraposition-double-negation-implication
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-contraposition-double-negation-implication-filled/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/synthetic-le-contra-double-neg-impli-finetuned-roberta-large-filled-bs-16/
(failed on bs=32)

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 20 \
  --output_dir Transformers/qqp/synthetic-logical-equivalence-finetuned-roberta-large-v4/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp-v4-bs-16
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/synthetic-logical-equivalence-finetuned-roberta-large-v4-bs-16/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp-v4-bs-16-positive-negative-1-2
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4-pos-neg-1-2/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/roberta-large-le-our-model-qqp-v4-pn-1-2/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp-v4-bs-16-positive-negative-1-3
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v4-pos-neg-1-3/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 16 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/roberta-large-le-our-model-qqp-v4-pn-1-3/

### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-our-model-v5/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 20 \
  --output_dir Transformers/qqp/synthetic-logical-equivalence-finetuned-roberta-large-v5/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-qqp-lreasoner
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-lreasoner/ \
  --task_name qqp \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/qqp/lreasoner-synthetic-logical-equivalence-finetuned-roberta-large/

##################################################################################################

### RoBERTa-large-cola
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path roberta-large \
  --task_name cola \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/cola/roberta-large/



### synthetic-logical-equivalence-finetuned-RoBERTa-large-cola
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large/ \
  --task_name cola \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/cola/synthetic-logical-equivalence-finetuned-roberta-large/

##################################################################################################

### RoBERTa-large-SST2
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path roberta-large \
  --task_name sst2 \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/sst2/roberta-large/



### synthetic-logical-equivalence-finetuned-RoBERTa-large-sst2
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large/ \
  --task_name sst2 \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/sst2/synthetic-logical-equivalence-finetuned-roberta-large/


### synthetic-logical-equivalence-finetuned-RoBERTa-large-sst2-lreasoner
python run_glue_no_trainer.py \
  --seed 42 \
  --model_name_or_path Transformers/roberta-large-lreasoner/ \
  --task_name sst2 \
  --max_length 256 \
  --per_device_train_batch_size 32 \
  --learning_rate 2e-5 \
  --num_train_epochs 10 \
  --output_dir Transformers/sst2/lreasoner-synthetic-logical-equivalence-finetuned-roberta-large/

##################################################################################################


export RECLOR_DIR=logiqa_data
export TASK_NAME=logiqa
export MODEL_NAME=roberta-large

CUDA_VISIBLE_DEVICES=6 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path roberta-large \
    --task_name logiqa \
    --do_train \
    --evaluate_during_training \
    --do_test \
    --do_lower_case \
    --data_dir $RECLOR_DIR \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 1e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/${MODEL_NAME} \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01


export RECLOR_DIR=reclor_data
export TASK_NAME=reclor
export MODEL_NAME=../amrlib-master/BERT/Transformers/roberta-large-lreasoner
export OUTPUT_DIR=lreasoner-logical-equivalence-reclor

CUDA_VISIBLE_DEVICES=6 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path ../amrlib-master/BERT/Transformers/roberta-large-lreasoner \
    --task_name reclor \
    --do_train \
    --evaluate_during_training \
    --do_test \
    --do_lower_case \
    --data_dir reclor_data \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 1e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/reclor/lreasoner-logical-equivalence-reclor \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01



CUDA_VISIBLE_DEVICES=3 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path ../amrlib-master/BERT/Transformers/roberta-large-lreasoner \
    --task_name logiqa \
    --do_train \
    --evaluate_during_training \
    --do_test \
    --do_lower_case \
    --data_dir logiqa_data \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 1e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/logiqa/lreasoner-logical-equivalence-reclor \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01

#############################################################################

CUDA_VISIBLE_DEVICES=5 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path roberta-large \
    --task_name pararule \
    --do_train \
    --evaluate_during_training \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 1e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/roberta-large-pararule \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01




CUDA_VISIBLE_DEVICES=6 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path ../amrlib-master/BERT/Transformers/roberta-large \
    --task_name pararule \
    --do_train \
    --evaluate_during_training \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 1e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/our-model-roberta-large-pararule \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01




CUDA_VISIBLE_DEVICES=7 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path ../amrlib-master/BERT/Transformers/roberta-large-lreasoner \
    --task_name pararule \
    --do_train \
    --evaluate_during_training \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 1e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/lreasoner-model-roberta-large-pararule \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01


######################################################################
CUDA_VISIBLE_DEVICES=1 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path roberta-large \
    --task_name pararule \
    --do_train \
    --evaluate_during_training \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 16   \
    --per_gpu_train_batch_size 16   \
    --gradient_accumulation_steps 1 \
    --learning_rate 1e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/roberta-large-pararule-bs-16 \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01


CUDA_VISIBLE_DEVICES=5 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path roberta-large \
    --task_name pararule \
    --do_train \
    --evaluate_during_training \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 16   \
    --per_gpu_train_batch_size 16   \
    --gradient_accumulation_steps 1 \
    --learning_rate 5e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/roberta-large-pararule-bs-16-lr-5e-5 \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01


CUDA_VISIBLE_DEVICES=6 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path roberta-large \
    --task_name pararule \
    --do_train \
    --evaluate_during_training \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 5e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/roberta-large-pararule-bs-8-lr-5e-5 \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01



#########################################################################
pararule depth-1 and reparaphrased testset

CUDA_VISIBLE_DEVICES=6 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path Checkpoints/lreasoner-model-roberta-large-pararule \
    --task_name pararule_depth_1 \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 5e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/lreasoner-model-roberta-large-pararule_depth_1 \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01

CUDA_VISIBLE_DEVICES=6 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path Checkpoints/lreasoner-model-roberta-large-pararule \
    --task_name pararule_depth_1_reparaphrased \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 5e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/lreasoner-model-roberta-large-pararule_depth_1_reparaphrased \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01

-----------------------------------------------------------------
CUDA_VISIBLE_DEVICES=5 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path Checkpoints/roberta-large-pararule \
    --task_name pararule_depth_1 \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 5e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/roberta-large-pararule_depth_1 \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01


CUDA_VISIBLE_DEVICES=5 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path Checkpoints/roberta-large-pararule \
    --task_name pararule_depth_1_reparaphrased \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 5e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/roberta-large-pararule_depth_1_reparaphrased \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01

--------------------------------------------------------------------
CUDA_VISIBLE_DEVICES=5 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path Checkpoints/our-model-roberta-large-pararule \
    --task_name pararule_depth_1 \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 5e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/our-model-roberta-large-pararule_depth_1 \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01


CUDA_VISIBLE_DEVICES=5 python run_multiple_choice.py \
    --model_type roberta \
    --model_name_or_path Checkpoints/our-model-roberta-large-pararule \
    --task_name pararule_depth_1_reparaphrased \
    --do_test \
    --do_lower_case \
    --data_dir pararule \
    --max_seq_length 256 \
    --per_gpu_eval_batch_size 8   \
    --per_gpu_train_batch_size 8   \
    --gradient_accumulation_steps 1 \
    --learning_rate 5e-05 \
    --num_train_epochs 10.0 \
    --output_dir Checkpoints/our-model-roberta-large-pararule_depth_1_reparaphrased \
    --fp16 \
    --logging_steps 200 \
    --save_steps 200 \
    --adam_betas "(0.9, 0.98)" \
    --adam_epsilon 1e-6 \
    --no_clip_grad_norm \
    --warmup_proportion 0.1 \
    --weight_decay 0.01