From 33390c4b74fa52875d6028c5c9aaf84f17288c25 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Sun, 9 Feb 2025 09:14:52 +0200 Subject: Use Q8_K_128 for IQ1_S_R4 and IQ1_M_R4 matrix multiplications (#194) * iq1_s_r4: Use Q8_K_128 instead of Q8_1_X4 for gemm (AVX2/Zen4) * iq1_m_r4: Use Q8_K_128 instead of Q8_1_X4 for gemm (AVX2/Zen4) * iq1_s_r4: Use Q8_K_128 instead of Q8_1_X4 for gemm (Neon) * iq1_m_r4: Use Q8_K_128 instead of Q8_0_X4 for gemm (Neon) * Simdify q8_K128 quantization also on Neon * Cleanup --------- Co-authored-by: Iwan Kawrakow --- ggml/include/ggml.h | 1 + 1 file changed, 1 insertion(+) (limited to 'ggml/include/ggml.h') diff --git a/ggml/include/ggml.h b/ggml/include/ggml.h index c307d42e..66bcb25a 100644 --- a/ggml/include/ggml.h +++ b/ggml/include/ggml.h @@ -415,6 +415,7 @@ extern "C" { GGML_TYPE_Q8_K16 = 147, GGML_TYPE_Q8_K32 = 148, GGML_TYPE_Q8_KR8 = 149, + GGML_TYPE_Q8_K128 = 150, GGML_TYPE_Q4_0_R8 = 202, GGML_TYPE_Q5_0_R4 = 206, -- cgit v1.2.3