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
In terms of code structure, to make the setup as simple as possible, all c++ extensions are in a separate package called pbt_C. We later call them from python code by importing that extra package. Nvidia used a similar approach: LINK
Because we want our lib to have no dependencies we compile all of the libraries statically and with fPIC flag
We also compile with a flag -D_GLIBCXX_USE_CXX11_ABI=0every c++ dependency because the libraries need the same version of basic_string as pytorch. For some reason pytorch is compiled with this flag set to 0 which means that we either need to recompile all dependencies the same way or we would need to compile pytorch from source. more details
A small note on compilers: Due to ABI versioning issues, the compiler you use to build your C++ extension must be ABI-compatible with the compiler PyTorch was built with.
Create an example method where you supply project-id, table name and columns and get a Pytorch Tensor with the results.
The text was updated successfully, but these errors were encountered: