Apex might be optional for who has difficulty installing this package?
"RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation" error occurred. My device is a 3090 GPU with pytorch 1.7.
I change scale = log_scale.mul_(0.5).tanh_().mul(self.alpha).add(1.0) to scale = log_scale.clone().mul_(0.5).tanh_().mul(self.alpha).add(1.0) and this error disappeared. Still not sure if it was correct.
Modifying tiny mistake in README.md.
Apex might be optional for who has difficulty installing this package?
"RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation" error occurred. My device is a 3090 GPU with pytorch 1.7. I change
scale = log_scale.mul_(0.5).tanh_().mul(self.alpha).add(1.0)toscale = log_scale.clone().mul_(0.5).tanh_().mul(self.alpha).add(1.0)and this error disappeared. Still not sure if it was correct.