| TOKENIZER=YOUR_PATH_TO_MODEL | |
| TOKENIZER_ID=roberta | |
| MODEL_PATH=YOUR_PATH_TO_MODEL | |
| python -m dense.driver.train --output_dir ../data/example/$TOKENIZER_ID/models \ | |
| --model_name_or_path $MODEL_PATH \ | |
| --tokenizer_name $TOKENIZER \ | |
| --save_steps 1000 \ | |
| --train_dir ../data/example/$TOKENIZER_ID/train \ | |
| --fp16 \ | |
| --per_device_train_batch_size 4 \ | |
| --learning_rate 5e-6 \ | |
| --num_train_epochs 10 \ | |
| --dataloader_num_workers 2 |