IQ2_XXS_R4 (#154)

* iq2_xxs_r4: Zen4

Disapointing gain: 134.7 t/s -> 151.1 t/s for PP-512
TG-128 is better: 3.45 -> 4.61 t/s @ 1 thread

* Minor

* iq2_xxs_r4: NEON

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2024-12-20 12:02:42 +01:00
committed by GitHub
parent 9904f8f691
commit 9dfd69bd93
10 changed files with 333 additions and 20 deletions

View File

@@ -418,6 +418,7 @@ extern "C" {
GGML_TYPE_Q4_K_R4 = 212,
GGML_TYPE_Q5_K_R4 = 213,
GGML_TYPE_Q6_K_R4 = 214,
GGML_TYPE_IQ2_XXS_R4= 216,
GGML_TYPE_IQ3_XXS_R4= 218,
GGML_TYPE_IQ4_NL_R4 = 220,
GGML_TYPE_IQ4_XS_R4 = 223,
@@ -495,8 +496,9 @@ extern "C" {
GGML_FTYPE_MOSTLY_Q2_K_R4 = 210, // except 1d tensors
GGML_FTYPE_MOSTLY_Q3_K_R4 = 211, // except 1d tensors
GGML_FTYPE_MOSTLY_Q4_K_R4 = 212, // except 1d tensors
GGML_FTYPE_MOSTLY_Q5_K_R4 = 215, // except 1d tensors
GGML_FTYPE_MOSTLY_Q5_K_R4 = 213, // except 1d tensors
GGML_FTYPE_MOSTLY_Q6_K_R4 = 214, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ2_XXS_R4= 215, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ3_XXS_R4= 217, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ4_NL_R4 = 219, // except 1d tensors
GGML_FTYPE_MOSTLY_IQ4_XS_R4 = 222, // except 1d tensors