CUDA: fix crash on large batch size for MoE models#13384

JohannesGaessler merged 1 commit intoggml-org:masterggml-org/llama.cpp:masterfrom

JohannesGaessler:cuda-fix-moe-max-ubJohannesGaessler/llama.cpp:cuda-fix-moe-max-ubCopy head branch name to clipboard