No standard values for scale_axis and predict_batch_size in trainer.py

[jg-qbig]

I am trying to train the model on English ptb but I am running into a problem when loading the inner model after the initial training run. The output of the train.py script is:

2024-08-02 13:34:24.472 | INFO     | __main__:cli_main:90 - Training the inside model!
Downloading: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████| 481/481 [00:00<00:00, 351kB/s]
Downloading: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████| 478M/478M [00:04<00:00, 115MB/s]
Global seed set to 42
GPU available: True, used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
Downloading: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████| 25.0/25.0 [00:00<00:00, 18.4kB/s]
Downloading: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████| 878k/878k [00:00<00:00, 10.3MB/s]
Downloading: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████| 446k/446k [00:00<00:00, 1.65MB/s]
Downloading: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████| 1.29M/1.29M [00:00<00:00, 6.72MB/s]
/opt/conda/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py:2203: LightningDeprecationWarning: `Trainer.gpus` was deprecated in v1.6 and will be removed in v1.8. Please use `Trainer.num_devices` or `Trainer.device_ids` to get device information instead.
  rank_zero_deprecation(
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [GPU-066a8ecc]
Epoch 4: 100%|██████████████████████████████████| 313/313 [03:52<00:00,  1.35it/s, loss=0.128, val_loss=0.264, val_accuracy=0.914, val_f1=0.926, val_mcc=0.829]
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [GPU-066a8ecc]
Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 63/63 [00:03<00:00, 16.91it/s]
───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
     Validate metric           DataLoader 0
───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
      val_accuracy          0.9144999980926514
         val_f1             0.9258777499198914
        val_loss            0.2638140022754669
         val_mcc            0.8288361430168152
───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
2024-08-02 13:39:22.960 | INFO     | __main__:cli_main:109 - Preparing data for self-training!
Traceback (most recent call last):
  File "/weakly-supervised-parsing/weakly_supervised_parser/train.py", line 182, in <module>
    cli_main()
  File "/weakly-supervised-parsing/weakly_supervised_parser/train.py", line 110, in cli_main
    train_self_trained, valid_self_trained = prepare_data_for_self_training(
TypeError: prepare_data_for_self_training() missing 2 required positional arguments: 'scale_axis' and 'predict_batch_size'

I used the standard arguments provided in the readme to start training:

python3 weakly-supervised-parsing/weakly_supervised_parser/train.py \
    --path_to_train_sentences ${TRAIN_SENTENCES_PATH} \
    --model_name_or_path roberta-base \
    --output_dir ${MODEL_PATH} \
    --max_epochs 10 \
    --lr 5e-6 \
    --train_batch_size 32 \
    --eval_batch_size 32 \
    --num_workers 16 \
    --inside_max_seq_length 256 \
    --outside_max_seq_length 64 \
    --num_labels 2 \
    --num_self_train_iterations 5 \
    --num_co_train_iterations 2 \
    --upper_threshold 0.995 \
    --lower_threshold 0.005 \
    --num_train_rows 100 \
    --num_valid_examples 100 \
    --seed 42

Digging further into the source code, the train.py file provides neither scale_axis nor predict_batch_size to the prepare_data_for_self_training_function call on line 110 although they are required arguments in the definition in prepare_data.py. Are scale_axis and predict_batch_size optional arguments? If not what values should they have?