Commit 398f00d65 for llama.cpp

commit 398f00d65769e73bb5cad57c809429b160f2bb0e
Author: Johannes Gäßler <johannesg@5d6.de>
Date:   Wed Nov 18 08:46:19 2026 +0100

    CUDA: use MMQ for MoE on V100 (#32504)

diff --git a/ggml/src/ggml-cuda/mmq.cu b/ggml/src/ggml-cuda/mmq.cu
index 2e6199766..c9bcbc2aa 100644
--- a/ggml/src/ggml-cuda/mmq.cu
+++ b/ggml/src/ggml-cuda/mmq.cu
@@ -460,7 +460,7 @@ bool ggml_cuda_should_use_mmq(enum ggml_type type, int cc, int64_t ne11, int64_t
 #endif //GGML_CUDA_FORCE_MMQ

     if (GGML_CUDA_CC_IS_NVIDIA(cc)) {
-        return !fp16_mma_hardware_available(cc) || ne11 < MMQ_DP4A_MAX_BATCH_SIZE;
+        return !fp16_mma_hardware_available(cc) || ne11 < MMQ_DP4A_MAX_BATCH_SIZE || n_experts > 0;
     }

     if (amd_mfma_available(cc)) {