forked from mrq/bitsandbytes-rocm
21 lines
328 B
Python
21 lines
328 B
Python
import bitsandbytes as bnb
|
|
import torch
|
|
|
|
p = torch.nn.Parameter(torch.rand(10,10).cuda())
|
|
a = torch.rand(10,10).cuda()
|
|
|
|
p1 = p.data.sum().item()
|
|
|
|
adam = bnb.optim.Adam([p])
|
|
|
|
out = a*p
|
|
loss = out.sum()
|
|
loss.backward()
|
|
adam.step()
|
|
|
|
p2 = p.data.sum().item()
|
|
|
|
assert p1 != p2
|
|
print('SUCCESS!')
|
|
print('Installation was successful!')
|