From 5d341757bc73efee0deba07a17679a965037753b Mon Sep 17 00:00:00 2001 From: Iwan Kawrakow Date: Mon, 29 Jul 2024 12:38:46 +0300 Subject: iq5_k: Basics Quantize/dequantize, CUDA dequantize --- ggml/include/ggml.h | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) (limited to 'ggml/include') diff --git a/ggml/include/ggml.h b/ggml/include/ggml.h index 2cb4af32..b7585ad6 100644 --- a/ggml/include/ggml.h +++ b/ggml/include/ggml.h @@ -389,8 +389,9 @@ extern "C" { GGML_TYPE_IQ1_BN = 34, GGML_TYPE_IQ2_BN = 35, GGML_TYPE_Q8_K64 = 36, - GGML_TYPE_IQ4_K = 37, - GGML_TYPE_IQ2_K = 38, + GGML_TYPE_IQ2_K = 37, + GGML_TYPE_IQ4_K = 38, + GGML_TYPE_IQ5_K = 39, GGML_TYPE_COUNT, }; @@ -437,8 +438,9 @@ extern "C" { GGML_FTYPE_MOSTLY_Q4_0_8_8 = 27, // except 1d tensors GGML_FTYPE_MOSTLY_IQ1_BN = 28, // except 1d tensors GGML_FTYPE_MOSTLY_IQ2_BN = 29, // except 1d tensors - GGML_FTYPE_MOSTLY_IQ4_K = 30, // except 1d tensors - GGML_FTYPE_MOSTLY_IQ2_K = 31, // except 1d tensors + GGML_FTYPE_MOSTLY_IQ2_K = 30, // except 1d tensors + GGML_FTYPE_MOSTLY_IQ4_K = 31, // except 1d tensors + GGML_FTYPE_MOSTLY_IQ5_K = 32, // except 1d tensors }; // available tensor operations: -- cgit v1.2.3