Missing pyproject.toml

Use a quantized LM or something smaller to make it easier to run the demo

Convert all argparse arguments to data classes + simple parsing

Assist in creating MLX implementations of UNet, NestedUnet, etc

Create separate dependency lists for cpu vs gpu, eg "pip install .[cpu]" vs "pip install .[gpu]"

Switch the default `pytest` call to run just CPU test cases

Separate python dependencies into optional groups

renaming instances of token_file to vocab_file

Web demo fails if generation is stopped before completion

Potential update to pyproject.toml to get web demos to run?