review

Author: TimDettmers

bitsandbytes/autograd/_functions.py

@@ -381,7 +381,7 @@ def backward(ctx, grad_output):
                 grad_A = F.mm_dequant(gradA32, SgradA32, SCgrad, state.SCBt).view(ctx.grad_shape).to(ctx.dtype_A)
 
             elif state.CB is not None:
-                CB = state.CB.to(ctx.dtype_A, copy=True).mul_(state.SCB.unsqueeze(1).div(127.0))
+                CB = state.CB.to(ctx.dtype_A, copy=True).mul_(state.SCB.unsqueeze(1).mul(1. / 127.0))
                 grad_A = torch.matmul(grad_output, CB).view(ctx.grad_shape).to(ctx.dtype_A)
             else:
                 raise Exception('State must contain either CBt or CB matrix for backward')