xflash96

Adding the saturate() function in the .cu should solve the problem. E.g., `inline __device__ float saturate(float x) { return x - (x1)*(1-x); }` This is cause by libtorch's versioning.

It seems that the CUDA module is not compiled. What version of PyTorch are you using? SATNet currently only supports pytorch==1.1.0. The CPP API changed in a later version, and...

I'm using the cudatoolkit that comes from PyTorch's official docker file: pytorch/pytorch:1.1.0-cuda10.0-cudnn7.5-devel. The "maxrregcount" argument is needed for older GPUs because NVCC may overspill the register if not set properly......

Confirmed. NVCC is required for custom CUDA extensions, and the "maxrrregcount" flag is also needed to work on Colab. For NVCC, it can be installed via $conda install -c conda-forge...

Thanks for your interest in the project! By the continuous nature of DL, the learnt clause matrix is not ternary. It is a mixture of (weighted) ternary clause matrices. So...

Sorry for the delayed reply. The "invalid on input" warning (satnet_cpp:194) means that there are Nan or Inf in the gradient, which didn't happen during our tests. Could you describe...

Sorry for the late update. I've updated the APIs to work with Pytorch:1.7.0. Also, I fixed the bug on the CPU version. May you confirm that it also works on...

Mostly it should be related to how PyTorch deal with the include_dirs in the CppExtension. What's the environment you're using? (Pytorch & Python version, the WSL docker tag).

It's mostly the path/lib problem for the NVCC toolchain on Windows. To try the package, I recommend using it on [Google Colab](https://colab.research.google.com/drive/1dRfepPLEE8N6BBZhXz8bbLDcPnRKaOcJ#forceEdit=true&offline=true&sandboxMode=true). If there is a necessity to run it...

Thanks for the check. It's been a few years but I remember this is because we changed the sign of dZ to save some computation... I.e., think the dz, U,...