ggml-cuda: tune RDNA3 Q6_K MMVQ nwarps (llama/23349)

diff --git a/ggml/src/ggml-cuda/mmvq.cu b/ggml/src/ggml-cuda/mmvq.cu
index da48f313a38b56ebd833536b068b19cdeb20b701..73a0991e20646d7c8d0e21fe274298ba6df9b85a 100644 (file)
--- a/ggml/src/ggml-cuda/mmvq.cu
+++ b/ggml/src/ggml-cuda/mmvq.cu
@@ -359,7 +359,9 @@ static constexpr __host__ __device__ int calc_nwarps(ggml_type type, int ncols_d
                 case GGML_TYPE_Q5_1:
                 case GGML_TYPE_Q8_0:
                 case GGML_TYPE_Q4_K:
+                    return 8;
                 case GGML_TYPE_Q6_K:
+                    return 2;
                 case GGML_TYPE_IQ4_NL:
                     return 8;
                 default: