cmake_windows_gpu job fails

[datumbox]

🐛 Describe the bug

The cmake_windows_gpu job on latest main started failing with:

-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.1 (found version "10.1") 
-- Caffe2: CUDA detected: 10.1
-- Caffe2: CUDA nvcc is: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.1/bin/nvcc.exe
-- Caffe2: CUDA toolkit directory: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.1
CMake Error at C:/tools/miniconda3/Lib/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:42 (message):
  PyTorch requires CUDA 10.2 or above.
Call Stack (most recent call first):
  C:/tools/miniconda3/Lib/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:88 (include)
  C:/tools/miniconda3/Lib/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package)
  CMakeLists.txt:17 (find_package)

It seems that despite defining CUDA10.2 below, something setups 10.1 on the paths:

vision/.circleci/config.yml

Lines 1704 to 1707 in c1ddb93

	- cmake_windows_gpu:
	cu_version: cu102
	name: cmake_windows_gpu
	python_version: '3.8'

This version mismatch was probably always there but the 10.1 version check and enforcement was probably introduced on PyTorch recently because this job passed yesterday.

Versions

latest main a2b4c65

[seemethere]

You need to upgrade this job to CUDA 10.2 were no longer supporting CUDA 10.1

[datumbox]

@seemethere My question was, shouldn't this be using already CUDA 10.2? See below:

vision/.circleci/config.yml

Lines 1704 to 1707 in c1ddb93

	- cmake_windows_gpu:
	cu_version: cu102
	name: cmake_windows_gpu
	python_version: '3.8'

I believe there is a bug on the current scripts that you use for installing the libraries. Could you please have a brief look as you are more familiar?