CUDA: fix matrix multiplication algorithm choice #14081

Job	Run time
Push Docker image to Docker Hub (light, .devops/llama-cli.Dockerfile, linux/amd64,linux/arm64)	9m 55s
Push Docker image to Docker Hub (server, .devops/llama-server.Dockerfile, linux/amd64,linux/arm64)	13m 18s
Push Docker image to Docker Hub (full, .devops/full.Dockerfile, linux/amd64,linux/arm64)	21m 34s
Push Docker image to Docker Hub (light-cuda, .devops/llama-cli-cuda.Dockerfile, linux/amd64)	35m 0s
Push Docker image to Docker Hub (server-cuda, .devops/llama-server-cuda.Dockerfile, linux/amd64)	36m 10s
Push Docker image to Docker Hub (full-cuda, .devops/full-cuda.Dockerfile, linux/amd64)	39m 55s
Push Docker image to Docker Hub (light-rocm, .devops/llama-cli-rocm.Dockerfile, linux/amd64,linux...	15m 9s
Push Docker image to Docker Hub (server-rocm, .devops/llama-server-rocm.Dockerfile, linux/amd64,l...	15m 25s
Push Docker image to Docker Hub (light-intel, .devops/llama-cli-intel.Dockerfile, linux/amd64)	7m 58s
Push Docker image to Docker Hub (server-intel, .devops/llama-server-intel.Dockerfile, linux/amd64)	8m 15s
	3h 22m 39s

Provide feedback