All the pretrained models' checkpoints are download from https://github.com/huggingface/transformers
Model command
Bert
python run_semeval.py
--model_type bert
--model_name_or_path bert-base-uncased
--task_name semeval
--do_train
--do_eval
--do_lower_case
--data_dir /root/code/data/semeval_data
--max_seq_length 64
--per_gpu_train_batch_size 32
--learning_rate 2e-5
--num_train_epochs 5.0
--output_dir /tmp/semeval_output/
Xlnet
python run_semeval.py
--model_type xlnet
--model_name_or_path xlnet-base-cased
--task_name semeval
--do_train
--do_eval
--do_lower_case
--data_dir /root/code/data/semeval_data
--max_seq_length 128
--per_gpu_train_batch_size 32
--learning_rate 2e-5
--num_train_epochs 5.0
--output_dir /tmp/outputs/semeval_xlnet_output/
Large-bert-uncased
python run_semeval.py
--model_type bert
--model_name_or_path /root/code/bert-large-uncased_L-24_H-1024_A-16
--task_name semeval
--do_train
--do_eval
--do_lower_case
--data_dir /root/code/data/semeval_data
--max_seq_length 64
--per_gpu_train_batch_size 32
--learning_rate 2e-5
--num_train_epochs 5.0
--output_dir /tmp/outputs/semeval_large_bert_output/
large-bert-cased
python run_semeval.py
--model_type bert
--model_name_or_path /root/code/bert-large-cased_L-24_H-1024_A-16
--task_name semeval
--do_train
--do_eval
--do_lower_case
--data_dir /root/code/data/semeval_data
--max_seq_length 64
--per_gpu_train_batch_size 32
--learning_rate 2e-5
--num_train_epochs 5.0
--output_dir /tmp/outputs/semeval_large_bert_cased_output/
xlm
python run_semeval.py
--model_type xlm
--model_name_or_path xlm-mlm-en-2048
--task_name semeval
--do_train
--do_eval
--do_lower_case
--data_dir /root/code/data/semeval_data
--max_seq_length 64
--per_gpu_train_batch_size 16
--learning_rate 2e-5
--num_train_epochs 5.0
--output_dir /tmp/outputs/semeval_xlm_output/
roberta
python run_semeval.py
--model_type roberta
--model_name_or_path roberta-base
--task_name semeval
--do_train
--do_eval
--do_lower_case
--data_dir /root/code/data/semeval_data
--max_seq_length 64
--per_gpu_train_batch_size 32
--learning_rate 2e-5
--num_train_epochs 8.0
--output_dir /tmp/outputs/semeval_roberta_output/
roberta-large-openai-detector
python run_semeval.py
--model_type roberta
--model_name_or_path roberta-large-openai-detector
--task_name semeval
--do_train
--do_eval
--do_lower_case
--data_dir /root/code/data/semeval_data
--max_seq_length 32
--per_gpu_train_batch_size 16
--learning_rate 2e-5
--num_train_epochs 5.0
--output_dir /tmp/outputs/semeval_roberta-large-openai-detector_output/
distilbert
python run_semeval.py
--model_type distilbert
--model_name_or_path distilbert-base-uncased
--task_name semeval
--do_train
--do_eval
--do_lower_case
--data_dir /root/code/data/semeval_data
--max_seq_length 64
--per_gpu_train_batch_size 32
--learning_rate 2e-5
--num_train_epochs 5.0
--output_dir /tmp/outputs/semeval_distilbert_output/
albert
python run_semeval.py
--model_type albert
--model_name_or_path /root/code/albert_v_xx_L12_H-4096_A-64
--task_name semeval
--do_train
--do_eval
--do_lower_case
--data_dir /root/code/data/semeval_data
--max_seq_length 64
--per_gpu_train_batch_size 32
--learning_rate 2e-5
--num_train_epochs 5.0
--output_dir /tmp/outputs/semeval_albert_output/
Xlm-roberta
python run_semeval.py
--model_type xlmroberta
--model_name_or_path xlm-roberta-large
--task_name semeval
--do_train
--do_eval
--do_lower_case
--data_dir /root/code/data/semeval_data
--max_seq_length 32
--per_gpu_train_batch_size 16
--learning_rate 2e-5
--num_train_epochs 5.0
--output_dir /tmp/outputs/semeval_xlmroberta_output/
flaubert
python run_semeval.py
--model_type flaubert
--model_name_or_path flaubert-large-cased
--task_name semeval
--do_train
--do_eval
--do_lower_case
--data_dir /root/code/data/semeval_data
--max_seq_length 32
--per_gpu_train_batch_size 16
--learning_rate 2e-5
--num_train_epochs 5.0
--output_dir /tmp/outputs/semeval_flaubert_output/