requirements.txt
python train.py --goal eval --device [your_device] --test_file_path [path_to_test_file] --weight_file_path [path_to_model]
python train.py
--source_lang ch #the input language
--target_lang fr #the output language
--goal train
--model_name google/mt5-base #the name of language model
--device cuda:0 #the device to ruin your code, cpu or cuda
--output_dir data/ #the folder to store the log and fine-tuned model
--max_token_num 216 #the number of tokens onput to the modle
--num_train_epochs 1000 #the number of training epoches
--train_file_path data/train.csv #the path to training sentence file
--test_file_path data/test.csv #the path to test sentence file
--batch_size 16 #the batch size of training and test
--lr 4e-5 #the learning rate
If you wanna use slurm cluster --job_name is to set the job name. If not, ignore it.