You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
After building both PyTorch and Apex from their master branches, my training script throws PyTorch C++ warnings at every training step. All warnings disappear when disabling Apex during training. Installing the official release of PyTorch 1.2 also seems to work without warnings.
/usr/local/lib/python3.5/dist-packages/torch/include/ATen/core/Tensor.h:297: UserWarning: Tensor.data<T>() is deprecated. Please use Tensor.data_ptr<T>() instead.
TORCH_WARN("Tensor.data<T>() is deprecated. Please use Tensor.data_ptr<T>() instead.");
PyTorch from master:f622ec8084c1778c6943474796efa52e9b9cf695 Apex from master:78c38db467c378759b4278a4ca8547763da5c91d CUDA Toolkit: 10.1 CuDNN: 7.1
The text was updated successfully, but these errors were encountered:
In general we've learned it's annoying to ask users to install a repo separate from Pytorch, especially one that relies on extension builds. I've been focused on getting Amp upstream, to avoid all this in the future. pytorch/pytorch#25081.
The API will be different, as requested by upstream, but more flexible and powerful imo.
After building both PyTorch and Apex from their master branches, my training script throws PyTorch C++ warnings at every training step. All warnings disappear when disabling Apex during training. Installing the official release of PyTorch 1.2 also seems to work without warnings.
PyTorch from master:
f622ec8084c1778c6943474796efa52e9b9cf695
Apex from master:
78c38db467c378759b4278a4ca8547763da5c91d
CUDA Toolkit: 10.1
CuDNN: 7.1
The text was updated successfully, but these errors were encountered: