AFT

- move preprocess word embeddings to dataset class - make word embeddings attribute of dataset class - decouple from datamodule logic - Should be identical to sentence embeddings logic -...

- Adapt Readme to new package structure - To do just before first release after development is finished

- Check current preprocessing implementation - Include further preprocessing steps if necessary

- make preprocessing possible for other languages -> see langdetect

Include Positional Invariance Layer. -> independent of model class. Optionally have submode invariance layers in between sub-layers.

Include pre-encoded column name embeddings and use as e.g. hadamard product between encodings and embeddings.

Include Jamba arch. -> Default to Attention Layer at beginning. @nwuestefeld

Include own mamba/mamba2 Triton version. @nwuestefeld