MaddoDev
Results
1
comments of
MaddoDev
trafficstars
For CUDA versions < 11.7 a target CUDA architecture must be explicitly provided via CUDA_DOCKER_ARCH
I had the same problem. Contributer from [this problem](https://github.com/ggerganov/llama.cpp/issues/5860#issuecomment-1988857249) helped `make CUDA_DOCKER_ARCH=sm_86` (for video cards from 3060 to 3090ti)