main usage: python -m distilbert_baseline --data /path/to/your.jsonl --mode train_or_test
Arguments:
-hor--help: show a help message and exit--data DATA: Path to .jsonl file containing dataset--mode MODE: which testing mode: 'train' or 'test'--use_class_weights USE_CLASS_WEIGHTS: use balanced class weights for training, default True--seed SEED: seed for random number generation, default 0xDEADBEEF--max_seq_len MAX_SEQ_LEN: maximum sequence length in transformer, default 128--batch_size BATCH_SIZE: training batch size, default 8--epochs EPOCHS: number of epochs of training, default 1--weight_decay WEIGHT_DECAY: AdamW weight decay, default 0.0--learning_rate LEARNING_RATE: AdamW learning rate, default 5e-5--adam_epsilon ADAM_EPSILON: AdamW epsilon, default 1e-8--warmup_steps WARMUP_STEPS: Warmup steps for learning rate schedule, default 0--max_grad_norm MAX_GRAD_NORM: max norm for gradient clipping, default 1.0