Noah Stier

Possible bug in half-precision scatter

4

I am getting different results for manually computing the mean of a half-precision tensor using `torch.mean` vs. `torch_scatter.scatter`. Is this expected behavior? ``` import torch import torch_scatter src_float = torch.randn(1_000_000).float()...

The drawer should have an incentive to choose a difficult word

1

I think there is a flaw in the game design: the drawer always scores the average of the guessers' scores, so the drawer has an incentive to choose the easiest...

Can @zmurez, or anyone else, suggest a configuration for training / fine-tuning Atlas that can learn acceptably well for experimentation using only a single GPU? I wonder if zmurez discovered...