From 4adb1d69d9f090eda7486e76feb43d4824803a4f Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Tue, 5 Dec 2023 18:15:51 +0200
Subject: [PATCH] cuda : add comment

---
 ggml-cuda.cu | 1 +
 1 file changed, 1 insertion(+)

diff --git a/ggml-cuda.cu b/ggml-cuda.cu
index 53e53a0d1634b..1200d1c888b42 100644
--- a/ggml-cuda.cu
+++ b/ggml-cuda.cu
@@ -7849,6 +7849,7 @@ static void ggml_cuda_mul_mat(const ggml_tensor * src0, const ggml_tensor * src1
 #endif // GGML_CUDA_FORCE_DMMV
 
             if (use_mul_mat_vec_q) {
+                // NOTE: this kernel does not support ggml_nrows(src1) > 1
                 ggml_cuda_op_mul_mat(src0, src1, dst, ggml_cuda_op_mul_mat_vec_q, true);
             } else {
                 ggml_cuda_op_mul_mat(src0, src1, dst, ggml_cuda_op_dequantize_mul_mat_vec, false);