#### bert srun --gres=gpu:1 python run_sentiment_classifier.py --do_train --do_eval --model_type bert_raw --model_name_or_path ../models/bert_base_uncased --task_name sst-2 --data_dir ../dataset/SST_2/ --num_train_epochs 3.0 --per_gpu_eval_batch_size 1000 --per_gpu_train_batch_size 32 --max_seq_length 128 --learning_rate 5e-5 --seed 11 #### xlnet python run_sentiment_classifier.py --do_train --do_eval --model_type xlnet_raw --model_name_or_path ../models/xlnet_base_cased --task_name sst-2 --data_dir ../dataset/SST_2/ --num_train_epochs 3.0 --per_gpu_eval_batch_size 1000 --per_gpu_train_batch_size 32 --max_seq_length 128 --learning_rate 2e-5 --seed 11 #### Roberta srun --gres=gpu:1 python run_sentiment_classifier.py --do_train --do_eval --model_type roberta_raw --model_name_or_path ../models/roberta_base_en --task_name sst-2 --data_dir ../dataset/SST_2/ --num_train_epochs 3.0 --per_gpu_eval_batch_size 1000 --per_gpu_train_batch_size 32 --max_seq_length 128 --learning_rate 2e-5 --seed 11 srun --gres=gpu:1 python run_sentiment_classifier.py --do_train --do_eval --model_type roberta_2task --model_name_or_path ../models/roberta_base_en --task_name sst-2 --data_dir ../dataset/SST_2/ --all_data_file ../dataset/SST_2/sst.binary.all --lexicon_file ../dataset/lexicon/SWN.word.polarity --num_train_epochs 3.0 --per_gpu_eval_batch_size 1000 --per_gpu_train_batch_size 32 --max_seq_length 128 --learning_rate 2e-5 --loss_type aggregation --loss_balance_type weight_sum --a 0.01 --b 0.0 --seed 11 srun --gres=gpu:1 python run_sentiment_classifier.py --do_train --do_eval --model_type roberta_2task --model_name_or_path ../models/roberta_base_en --task_name sst-2 --data_dir ../dataset/SST_2/ --all_data_file ../dataset/SST_2/sst.binary.all --lexicon_file ../dataset/lexicon/SWN.word.polarity --num_train_epochs 3.0 --per_gpu_eval_batch_size 1000 --per_gpu_train_batch_size 32 --max_seq_length 128 --learning_rate 2e-5 --loss_type aggregation --loss_balance_type add_vec --c 1.0 --seed 11 # ===================================================================== ### SentiLARE pretrained roberta model+SST-2 (this is baseline) srun --gres=gpu:1 python run_sentiment_classifier.py --do_train --do_eval --model_type roberta_raw --model_name_or_path ../models/SentiLARE_pretrain_roberta --task_name sst-2 --data_dir ../dataset/SST_2/ --num_train_epochs 3.0 --per_gpu_eval_batch_size 1000 --per_gpu_train_batch_size 32 --max_seq_length 128 --learning_rate 2e-5 --seed 11 srun --gres=gpu:1 python run_sentiment_classifier.py --do_train --do_eval --model_type roberta_2task --model_name_or_path ../models/SentiLARE_pretrain_roberta --task_name sst-2 --data_dir ../dataset/SST_2/ --all_data_file ../dataset/SST_2/sst.binary.all --lexicon_file ../dataset/lexicon/SWN.word.polarity --num_train_epochs 3.0 --per_gpu_eval_batch_size 1000 --per_gpu_train_batch_size 32 --max_seq_length 128 --learning_rate 2e-5 --loss_type joint --loss_balance_type weight_sum --a 0.01 --b 0.0 --seed 11 srun --gres=gpu:1 python run_sentiment_classifier.py --do_train --do_eval --model_type roberta_2task --model_name_or_path ../models/SentiLARE_pretrain_roberta --task_name sst-2 --data_dir ../dataset/SST_2/ --all_data_file ../dataset/SST_2/sst.binary.all --lexicon_file ../dataset/lexicon/SWN.word.polarity --num_train_epochs 3.0 --per_gpu_eval_batch_size 1000 --per_gpu_train_batch_size 32 --max_seq_length 128 --learning_rate 2e-5 --loss_type aggregation --loss_balance_type add_vec --c 0.01 --seed 11