From 6421d2c1f2723e93d5989f5f2d7990dd50a50a16 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Johannes=20G=C3=A4=C3=9Fler?= Date: Sun, 4 Oct 2026 14:10:43 +0200 Subject: [PATCH] CUDA: fix MMQ memory fault if n_expert >> n_ubatch (llama/29941) --- ggml/src/ggml-cuda/mmq.cu | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/ggml/src/ggml-cuda/mmq.cu b/ggml/src/ggml-cuda/mmq.cu index 3e1721bbe..43603bb8a 100644 --- a/ggml/src/ggml-cuda/mmq.cu +++ b/ggml/src/ggml-cuda/mmq.cu @@ -255,7 +255,7 @@ void ggml_cuda_mul_mat_q( } const size_t nbytes_src1_q8_1 = ne12*n_expert_used*ne10_padded * y_block_size/y_values_per_block + - ggml_cuda_mmq_get_J_max(src0->type, fallback, cc, ne11) * sizeof(block_q8_1_mmq); + ggml_cuda_mmq_get_J_max(src0->type, fallback, cc, ne12) * sizeof(block_q8_1_mmq); ggml_cuda_pool_alloc src1_q8_1(ctx.pool(), nbytes_src1_q8_1); ggml_cuda_pool_alloc src1_scale(ctx.pool()); if (src0->type == GGML_TYPE_NVFP4 && use_native_fp4) {