Usage (seecli.py):

python cli.py fit ...
python cli.py predict ...

• use TrainingConfig in utils.py to set (hyper)parameters
• dataset.py - dataset wrapper implementation, the expected data format is described in MultiModalDataset. MultiModalTestDataset may be useful for batch inference, not really needed for online inference
• MultiModalClassifier in model.py implements the architecture itself. Metrics and a loss function are also defined there
• cli.py implements a training loop and offline prediction. Note that LabelEncoder and MinMaxScaler used for tabular data processing are also pickled to be reused on inference.

Not intended to be as a go-to solution, but rather a comprehensive example of a custom multimodal model that handles both text and tabular data simultaneously. In Pytorch-Lightning