From 7f61b3068e18728e5e7e2b95546ff03dd2fd41ac Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Thu, 6 Feb 2025 14:08:52 +0200 Subject: IQ1_M_R4: better 1.75 bpw quants (#187) * iq1_m_r4: basics (quantize/dequantize) * iq1_m_r4: Zen4 gemm * iq1_m_r4: neon gemm * iq1_m_r4: switch to q8_0_x4 also on AVX2/Zen4 With the deltas being per group of 8, we cannot make use of the q8 sums stored in q8_1, so we get a tiny gain by using q8_0_x4. * iq1_m_r4: rename mul_mat_iq1_m_r4_q8_1 to mul_mat_iq1_m_r4_q8_0 --------- Co-authored-by: Iwan Kawrakow --- ggml/include/ggml.h | 2 ++ 1 file changed, 2 insertions(+) (limited to 'ggml/include/ggml.h') diff --git a/ggml/include/ggml.h b/ggml/include/ggml.h index 9668dc32..77ac33a9 100644 --- a/ggml/include/ggml.h +++ b/ggml/include/ggml.h @@ -432,6 +432,7 @@ extern "C" { GGML_TYPE_IQ3_S_R4 = 221, GGML_TYPE_IQ2_S_R4 = 222, GGML_TYPE_IQ4_XS_R4 = 223, + GGML_TYPE_IQ1_M_R4 = 229, GGML_TYPE_BF16_R16 = 230, GGML_TYPE_Q6_0_R4 = 233, GGML_TYPE_IQ2_BN_R4 = 335, @@ -516,6 +517,7 @@ extern "C" { GGML_FTYPE_MOSTLY_IQ3_S_R4 = 220, // except 1d tensors GGML_FTYPE_MOSTLY_IQ2_S_R4 = 221, // except 1d tensors GGML_FTYPE_MOSTLY_IQ4_XS_R4 = 222, // except 1d tensors + GGML_FTYPE_MOSTLY_IQ1_M_R4 = 223, // except 1d tensors GGML_FTYPE_MOSTLY_BF16_R16 = 224, // except 1d tensors GGML_FTYPE_MOSTLY_Q6_0_R4 = 227, // except 1d tensors GGML_FTYPE_MOSTLY_IQ2_BN_R4 = 329, // except 1d tensors -- cgit v1.2.3