diff --git a/ggml/src/ggml-cuda/mmq.cuh b/ggml/src/ggml-cuda/mmq.cuh index 15af470843..4c74f0bfba 100644 --- a/ggml/src/ggml-cuda/mmq.cuh +++ b/ggml/src/ggml-cuda/mmq.cuh @@ -52,6 +52,7 @@ struct block_fp4_mmq { static_assert(sizeof(block_q8_1_mmq) == 4*QK8_1 + 4*sizeof(half2), "Unexpected block_q8_1_mmq size"); static_assert(sizeof(block_q8_1_mmq) == 4*sizeof(block_q8_1), "Unexpected block_q8_1_mmq size"); +static_assert(sizeof(block_fp4_mmq) == 72, "Unexpected block_fp4_mmq size"); static mmq_q8_1_ds_layout mmq_get_q8_1_ds_layout(const ggml_type type_x) { switch (type_x) {