| 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!') |