emaad

@fnzhan any updates? Really interested to see if there are any key improvements.

> @afiaka87 yup, should be doable! let me think about the interface - perhaps something like `--vqvae-path` and `--vqvae-config-path` > > or perhaps, to stay simple, i should just default...

@Lamikins I believe the training issues come from an error in the masking formula. I've ammended the error: https://github.com/dome272/MaskGIT-pytorch/pull/16. @xuesongnie

@xuesongnie it's because the mask calculated is applied to the wrong values. The other option would be to do ```r = math.floor(1-self.gamma(np.random.uniform()) * z_indices.shape[1])```, but I don't like that because...