mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-03-14 15:57:37 +00:00
Fix q4_1 and q5_1 on Arm (#348)
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
@@ -741,7 +741,11 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||||||
.from_float_ref = (ggml_from_float_t) quantize_row_q4_1_ref,
|
.from_float_ref = (ggml_from_float_t) quantize_row_q4_1_ref,
|
||||||
.vec_dot = ggml_vec_dot_q4_1_q8_1,
|
.vec_dot = ggml_vec_dot_q4_1_q8_1,
|
||||||
#if GGML_USE_IQK_MULMAT
|
#if GGML_USE_IQK_MULMAT
|
||||||
|
#if defined __AVX2__
|
||||||
.vec_dot_type = GGML_TYPE_Q8_2_X4,
|
.vec_dot_type = GGML_TYPE_Q8_2_X4,
|
||||||
|
#else
|
||||||
|
.vec_dot_type = GGML_TYPE_Q8_1_X4,
|
||||||
|
#endif
|
||||||
#else
|
#else
|
||||||
.vec_dot_type = GGML_TYPE_Q8_1,
|
.vec_dot_type = GGML_TYPE_Q8_1,
|
||||||
#endif
|
#endif
|
||||||
@@ -809,7 +813,11 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||||||
.from_float_ref = (ggml_from_float_t) quantize_row_q5_1_ref,
|
.from_float_ref = (ggml_from_float_t) quantize_row_q5_1_ref,
|
||||||
.vec_dot = ggml_vec_dot_q5_1_q8_1,
|
.vec_dot = ggml_vec_dot_q5_1_q8_1,
|
||||||
#if GGML_USE_IQK_MULMAT
|
#if GGML_USE_IQK_MULMAT
|
||||||
|
#ifdef __AVX2__
|
||||||
.vec_dot_type = GGML_TYPE_Q8_2_X4,
|
.vec_dot_type = GGML_TYPE_Q8_2_X4,
|
||||||
|
#else
|
||||||
|
.vec_dot_type = GGML_TYPE_Q8_1_X4,
|
||||||
|
#endif
|
||||||
#else
|
#else
|
||||||
.vec_dot_type = GGML_TYPE_Q8_1,
|
.vec_dot_type = GGML_TYPE_Q8_1,
|
||||||
#endif
|
#endif
|
||||||
|
|||||||
Reference in New Issue
Block a user