From 6e5156cab5c6d2858e1ecd5bc4dc5db81c71de39 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Tue, 1 Apr 2025 08:29:25 +0200 Subject: Fix #300 (#301) Co-authored-by: Iwan Kawrakow --- ggml/src/iqk/iqk_mul_mat.cpp | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) (limited to 'ggml/src') diff --git a/ggml/src/iqk/iqk_mul_mat.cpp b/ggml/src/iqk/iqk_mul_mat.cpp index cf512ba5..1c8a991d 100644 --- a/ggml/src/iqk/iqk_mul_mat.cpp +++ b/ggml/src/iqk/iqk_mul_mat.cpp @@ -15514,8 +15514,13 @@ struct HelperQ8KVR8 : public BaseHelper { template struct HelperQ40 final : public BaseHelper { using Base = BaseHelper; +#if defined __AVX2__ using block_q8 = block_q8_2; constexpr static int block_size_q = QK8_2; +#else + using block_q8 = block_q8_0; + constexpr static int block_size_q = QK8_0; +#endif HelperQ40(const char * data, int stride) : Base(data, stride) {} // Needed for v * softmax(k * q) @@ -15558,8 +15563,8 @@ struct HelperQ40 final : public BaseHelper { template struct HelperQ41 final : public BaseHelper { using Base = BaseHelper; - using block_q8 = block_q8_1; - constexpr static int block_size_q = QK8_1; + using block_q8 = block_q8_2; + constexpr static int block_size_q = QK8_2; HelperQ41(const char * data, int stride) : Base(data, stride) {} // Needed for v * softmax(k * q) @@ -16414,7 +16419,7 @@ struct FlashQKfp32 { #ifdef __aarch64__ MAKE_FUNCS(mul_mat_qX_0_q8_0>) { -- cgit v1.2.3