Jamie Bloxham

Initialize Models on GPU

1

This might not be easily possible. In PyTorch, `nn.module` instances always seem to get initialized on the CPU, not the GPU, even though it is possible to initialize tensors directly...

[Efficient SAM](https://openreview.net/forum?id=n0OeTdNRG0Q) (anonymous, 2022) is a proposed duo of SAM optimizations to reduce the throughput hit of SAM. The composer repo already supports an `interval` hyperparameter which has empirically been...

These files still reference methods from last year.