Henry Mao

Seems like there are problems with Jest and Enzyme :( https://github.com/facebook/jest/issues/1175

Running into the same issue

Getting the same issue here as @Stvad. Seems like web extension production build is not robust (it works well in `parcel watch` dev mode)? I followed the changes from @Kjoep...

Running into the same issue during `parcel build`. `--no-content-hash` fixes it.

Using the method proposed by @Thomas214 definitely speed things up, but as a library I believe Catalyst should handle this in the backend by caching data in an easy-to-access format.

Seems like with numba right now it's possible to use FP16 inputs, but not FP16 shared memory. Does full fp16 support in CUDA mean we can use fp16 shared memory?...

It's a generation-level feature. You don't have to train it on interpolation. We have a paper written on the details that we may release in the future.

@DefinitlyEvil Yes, definitely! It's a difficult research task though which will require some smart solution to get working.

@DefinitlyEvil Discord Calclavia#9037 It's not as simple as just multiple networks, because each instrument needs to be conditioned on another instrument for harmony. It'll have to be done somewhat efficiently.

@yyht A few questions: - Did you initialize \alpha to zero? - How did you initialize the embedding matrix? We found that GPT2's embedding initialization doesn't work very well.