CUDA: fix crash on large batch size for quant. MoE#13537

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

JohannesGaessler:cuda-fix-dsr1-max-batchJohannesGaessler/llama.cpp:cuda-fix-dsr1-max-batchCopy head branch name to clipboard