diff --git a/ggml/src/iqk/iqk_mul_mat.cpp b/ggml/src/iqk/iqk_mul_mat.cpp index 2c4987a0..ed7309cd 100644 --- a/ggml/src/iqk/iqk_mul_mat.cpp +++ b/ggml/src/iqk/iqk_mul_mat.cpp @@ -15315,9 +15315,11 @@ struct HelperIQ4nl final : public BaseHelper { #ifdef __aarch64__ using block_q8 = block_q8_0; HelperIQ4nl(const char * data, int stride) : Base(data, stride), values(vld1q_s8(iq4k_values)) {} + constexpr static int block_size_q = QK8_0; #else HelperIQ4nl(const char * data, int stride) : Base(data, stride) {} using block_q8 = block_q8_1; + constexpr static int block_size_q = QK8_1; #endif // Needed for v * softmax(k * q)