From b30c9e10d8710a49b2d2ab98d086b9f11bfaa228 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Wed, 9 Oct 2024 12:54:40 +0300 Subject: New SOTA quantization: 4.25 bpw IQ4_KS (#83) * iq4_k_xxs: basics * WIP + adding iq3_kl quantization mix * iq4_xxs: this looks very viable compared to iq4_xs At the same 4.25 bpw PPL is always better, for some models significantly better. I'll rename to iq4_ks and keep it. * iq4_xxs: CUDA dot product We get TG-128 = 126 t/s for LLaMA-3.1-8B, compared to 123 t/s for q4_0. * iq4_xxs: scalar CPU dot product Also fix the breakage I caused with the dedicated work buffer quantization portion when the multiplication is not done via iqk_mul_mat. * iq4_xxs: Zen4 I noticed that iq4_xs is wrong on Zen4 (and possibly AVX2). Again the same mistake of packing int32_t back to int16_t, which overflows occasionally (just occasionally, that's why the result doesn't look completely wrong, so I didn't notice). * Fix iq4_xs (Zen4) * iq4_xxs: AVX2 * iq4_xxs: ARM_NEON * iq4_xxs: Metal * iq4_xxs: slightly faster TG on Metal * iq4_xxs: rename to iq4_ks After all, tt is a smaller variant of iq4_k. * iq3_kl: use iq4_ks instead of iq4_k/iq4_xs --------- Co-authored-by: Iwan Kawrakow --- ggml/src/ggml-quants.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'ggml/src/ggml-quants.c') diff --git a/ggml/src/ggml-quants.c b/ggml/src/ggml-quants.c index f5fff22e..40978ac0 100644 --- a/ggml/src/ggml-quants.c +++ b/ggml/src/ggml-quants.c @@ -14947,7 +14947,7 @@ bool ggml_validate_row_data(enum ggml_type type, const void * data, size_t nbyte return false; } - if (type != GGML_TYPE_IQ2_TN && type != GGML_TYPE_IQ1_TN && nbytes % ggml_type_size(type) != 0) { + if (type != GGML_TYPE_IQ2_TN && type != GGML_TYPE_IQ1_TN && type != GGML_TYPE_IQ4_KS && nbytes % ggml_type_size(type) != 0) { fprintf(stderr, "%s: invalid size %zu for type %s (type size = %zu)\n", __func__, nbytes, ggml_type_name(type), ggml_type_size(type)); return false; } @@ -15166,6 +15166,7 @@ bool ggml_validate_row_data(enum ggml_type type, const void * data, size_t nbyte case GGML_TYPE_IQ6_K: break; case GGML_TYPE_IQ2_TN: break; case GGML_TYPE_IQ1_TN: break; + case GGML_TYPE_IQ4_KS: break; case GGML_TYPE_Q4_0_4_4: case GGML_TYPE_Q4_0_4_8: { -- cgit v1.2.3