From 2700d3af36e5ce11f6c2c9644130635f83bae6d5 Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Thu, 12 Dec 2024 16:04:20 +0100 Subject: IQ4_K_R4 (#138) * iq4_k_r4: WIP * iq4_k_r4: Zen4 and hopefully AVX2 On Zen4 we get PP-512(LLaMA-3.1-8B) = 232.6 t/s, up from 182.2 t/s for iq4_k. Applying the extra shift costs a ~6 performance penalty. * iq4_k_r4: AVX2 PP-512 = 227.60 t/s. The shifts are really costly. * iq4_k_r4: NEON We get PP-512(LLaMA-3.1-8B) = 108 t/s, up from 58.2 t/s for iq4_k. --------- Co-authored-by: Iwan Kawrakow --- ggml/src/ggml-quants.c | 1 + 1 file changed, 1 insertion(+) (limited to 'ggml/src/ggml-quants.c') diff --git a/ggml/src/ggml-quants.c b/ggml/src/ggml-quants.c index ff857087..64bd9459 100644 --- a/ggml/src/ggml-quants.c +++ b/ggml/src/ggml-quants.c @@ -15207,6 +15207,7 @@ bool ggml_validate_row_data(enum ggml_type type, const void * data, size_t nbyte case GGML_TYPE_Q4_K_R4: break; case GGML_TYPE_Q5_K_R4: break; case GGML_TYPE_Q6_K_R4: break; + case GGML_TYPE_IQ4_K_R4: break; case GGML_TYPE_Q4_0_4_4: case GGML_TYPE_Q4_0_4_8: { -- cgit v1.2.3