| // `torch::kFloat8_e4m3fn` is supported since PyTorch 2.1 | |
| // `cuTensorMapEncodeTiled` is supported since CUDA Driver API 12.1 | |
| // `cublasGetErrorString` is supported since CUDA Runtime API 11.4.2 | |
| // `CUBLASLT_MATMUL_DESC_FAST_ACCUM` and `CUBLASLT_MATMUL_DESC_SM_COUNT_TARGET` are supported since CUDA Runtime API 11.8 | |