CUDA: fix Gemma 2 numerical issues for FA (#9166) #14393

Job	Run time
Push Docker image to Docker Hub (light, .devops/llama-cli.Dockerfile, linux/amd64,linux/arm64)	13m 41s
Push Docker image to Docker Hub (server, .devops/llama-server.Dockerfile, linux/amd64,linux/arm64)	18m 49s
Push Docker image to Docker Hub (full, .devops/full.Dockerfile, linux/amd64,linux/arm64)	33m 54s
Push Docker image to Docker Hub (light-cuda, .devops/llama-cli-cuda.Dockerfile, linux/amd64)	1h 33m 39s
Push Docker image to Docker Hub (server-cuda, .devops/llama-server-cuda.Dockerfile, linux/amd64)	1h 30m 36s
Push Docker image to Docker Hub (full-cuda, .devops/full-cuda.Dockerfile, linux/amd64)	1h 38m 10s
Push Docker image to Docker Hub (light-rocm, .devops/llama-cli-rocm.Dockerfile, linux/amd64,linux...	38m 11s
Push Docker image to Docker Hub (server-rocm, .devops/llama-server-rocm.Dockerfile, linux/amd64,l...	37m 35s
Push Docker image to Docker Hub (light-intel, .devops/llama-cli-intel.Dockerfile, linux/amd64)	16m 15s
Push Docker image to Docker Hub (server-intel, .devops/llama-server-intel.Dockerfile, linux/amd64)	15m 3s
	7h 35m 53s

Provide feedback