(dev-util/nvidia-cuda-toolkit-12.5.0-r1:0/12.5.0::gentoo, ebuild scheduled for merge) USE="examples nsight profiler sanitizer vis-profiler -debugger -rdma" ABI_X86="(64)" conflicts with <dev-util/nvidia-cuda-toolkit-12.4.0:=[profiler] required by (sci-libs/caffe2-2.3.0-r3:0/0::gentoo, installed) USE="cuda distributed fbgemm ffmpeg gloo mpi nnpack numpy onednn openblas opencl opencv openmp qnnpack xnnpack -flash -mkl (-rocm)" ABI_X86="(64)" AMDGPU_TARGETS="gfx1030 gfx906 gfx908 gfx90a -gfx1010 -gfx1011 -gfx1012 -gfx1031 -gfx1100 -gfx1101 -gfx1102 -gfx803 -gfx900" PYTHON_SINGLE_TARGET="python3_11 -python3_10 -python3_12" ^ ^^^^^^ <dev-util/nvidia-cuda-toolkit-12.4.0:0/12.3.2=[profiler] required by (sci-libs/caffe2-2.3.0-r3:0/0::gentoo, installed) USE="cuda distributed fbgemm ffmpeg gloo mpi nnpack numpy onednn openblas opencl opencv openmp qnnpack xnnpack -flash -mkl (-rocm)" ABI_X86="(64)" AMDGPU_TARGETS="gfx1030 gfx906 gfx908 gfx90a -gfx1010 -gfx1011 -gfx1012 -gfx1031 -gfx1100 -gfx1101 -gfx1102 -gfx803 -gfx900" PYTHON_SINGLE_TARGET="python3_11 -python3_10 -python3_12" ^ ^^^^^^^^^^^^^^^^ Reproducible: Always
https://pytorch.org/get-started/locally/ caffe2-2.3.x doesn't support cuda-12.4 per their support matrix, most likely due to the torch fubar cuda had. They also don't list 12.5. The upcoming caffe-2.4.0 release might support it. Do you have any info that it should support cuda-12.5?
ok thanks, that explains this, so its not a gentoo problem but upstream...