Quality of Life improvements for pytorch
Basically a combination of monkey-patches and additional functionality which is missing from pytorch. Use at your own peril :o
import fix_torch
import torch
x = torch.randn(13,127,12,3)
y = torch.randn(13,127,3,17)
v = torch.randn(8,4,4)
z = x@y # do bmm without torch being annoying about it
tr = torch.trace(v) # do trace on batched matrices
w = x.reshape(...,4,3,3)