Sasha Rush

Turn the notebooks into scripts to compute final eval scores.

Note what it requires. For instance needs at least torch-hdf5 and python hdf5. Even better we can do this with a clear error. For instance in Elements RNN they do:...

Let's add a bash script to generate all the results and timing numbers. This is helpful for people trying to improve on the current results.

HuggingFace Hub

3

It seems like HuggingFace is only supported for local models. It would be nice to support calling model on the Hub with the InferenceAPI. I think this is relatively straightforward...

Problems with non-Llama

2

I get this error with other models: `AssertionError: error: target_token_ids {28705} not unique to target strings {' Negative.', ' Positive.'}` I'll see if I can debug it.

Autoplay sliders would be really fun.

Support numpy/jax

1

My summer project is to port all the layout code to use Numpy. This mainly will mean rewriting V2, P2, Affine, colors to be in matrix representation. * I think...