What exactly is --train-split used for (when llm training)?

[heikomat]

In the autotrain-advanced video at 11:20 Abhishek states that for llm training, the name of the csv-file with the training data is the train-split.

So i thought that i could name my file training-data.csv and supply --train-split training-data.
This DID create a autotrain-folder with some files in it, but it then crashes with this error:

  File "/Users/heiko/.pyenv/versions/3.10.13/lib/python3.10/site-packages/autotrain/trainers/common.py", line 87, in wrapper
    return func(*args, **kwargs)
  File "/Users/heiko/.pyenv/versions/3.10.13/lib/python3.10/site-packages/autotrain/trainers/clm/__main__.py", line 105, in train
    train_data, valid_data = process_input_data(config)
  File "/Users/heiko/.pyenv/versions/3.10.13/lib/python3.10/site-packages/autotrain/trainers/clm/__main__.py", line 42, in process_input_data
    train_data = load_from_disk(config.data_path)[config.train_split]
  File "/Users/heiko/.pyenv/versions/3.10.13/lib/python3.10/site-packages/datasets/dataset_dict.py", line 59, in __getitem__
    return super().__getitem__(k)
KeyError: 'training-data'

the only way i could start autotrain was by not providing --train-split and naming my file train.csv.

I'm not sure if i misunderstand what --train-split is supposed to do, or if it is broken somehow.
It seems to currently be impossible to train an llm when providing --train-split. Is that expected behaviour?