Libtorch v2.7.1

[dfalbel]

Updates for LibTorch v2.7.1

On Ci when running CUDA 12.8 we see:

CUDA error: no kernel image is available for execution on the device
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1

This is likely because the our GPU NVIDIA 1080Ti uses Pascal architecture (6.1) and that's now not supported in 12.8. See:

TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"
case ${CUDA_VERSION} in
    12.8)
        TORCH_CUDA_ARCH_LIST="7.5;8.0;8.6;9.0;10.0;12.0+PTX" #removing sm_50-sm_70 as these architectures are deprecated in CUDA 12.8 and will be removed in future releases
        EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
        ;;
    12.6)
        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"
        EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
        ;;
    12.4)
        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"
        EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
        ;;
    11.8)
        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};3.7;9.0"
        EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
        ;;
    *)
        echo "unknown cuda version $CUDA_VERSION"
        exit 1
        ;;
esac

in the repo

Thus CI now uses 12.6. Looking forward we'll need to upgrade our runner GPU or move to using a different CI runner.