bytecode-pre-trained

1. unzip data files (1. test_files.zip 2. train.zip 3. valid.zip)

2. generate input paths by using generate_input_path.py

• you should put configurations on generate_input_path.py
• directory path and output text file name
• maximum number is 1) train = 40,000 2) validation = 8,000 3) test = 12,000

3. Put your environment on pretrain_byteT5.py and pretrain_byteBERT.py

• train.txt path
• valid.txt path
• test.txt path
• GPU environment
• save point directory

4. Run byteT5.py and byteBERT.py with train mode

• use python3 Model/byteT5.py --mode train

5. Run byteT5.py and byteBERT.py with test mode

• use python3 Model/byteT5.py --mode test --model_file "your savepoint path"