installation error

[adeclaire18]

Cannot install successfully.

Here are the error output:

• 1 error detected in the compliation of "/tmp/tmpxft_00000f76_00000000-6_rectify_cuda_cpp1.ii".
• error: command '.usr/local/cuda/bin/nvcc' failed with exit status 1

From my perspective, it occur error during compiling gpu part. I even tried to change PATH_TO_CUDA, but the SAME error are STILL there.

Here are my environments:

• torch: 1.4.0
• torchvision: 0.5.0
• nvcc -V: 10.1
• gcc & g++: 5

I have tried many times, checking package versions, pull from source, create new conda envs, uninstalls. BUT STILL ERROR!

Have anyone encountered the same problem? What should I do?

[adeclaire18]

Good news. inspired by another issue #161.

1. Download source file from: https://pypi.org/project/torch-encoding/#files
2. "pip install " a few dependencies
3. Then installation succeed.

But I get new errors while training: "gpu" has no attribute 'expectation_forward'.