summaryrefslogtreecommitdiff
path: root/ggml/src/ggml.c
diff options
context:
space:
mode:
Diffstat (limited to 'ggml/src/ggml.c')
-rw-r--r--ggml/src/ggml.c24
1 files changed, 23 insertions, 1 deletions
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index c3cda4c4..6bfeca1e 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -11,6 +11,7 @@
#include "ggml-quants.h"
#include "ggml.h"
#include "ggml-aarch64.h"
+#include "iqk/iqk_quantize.h"
#if GGML_USE_IQK_MULMAT
#include "iqk/iqk_mul_mat.h"
#endif
@@ -978,7 +979,19 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
.ncols = 8,
.gemv = ggml_gemv_q4_0_8x8_q8_0,
.gemm = ggml_gemm_q4_0_8x8_q8_0,
- }
+ },
+ [GGML_TYPE_IQ4_K] = {
+ .type_name = "iq4_k",
+ .blck_size = QK_K,
+ .type_size = sizeof(block_iq4_k),
+ .is_quantized = true,
+ .to_float = (ggml_to_float_t) dequantize_row_iq4_k,
+ .from_float = quantize_row_iq4_k,
+ .from_float_ref = (ggml_from_float_t)quantize_row_iq4_k_ref,
+ .vec_dot = vec_dot_iq4_k_q8_k,
+ .vec_dot_type = GGML_TYPE_Q8_K,
+ .nrows = 1,
+ },
};
// For internal test use
@@ -3328,6 +3341,7 @@ enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype) {
case GGML_FTYPE_MOSTLY_IQ2_BN: wtype = GGML_TYPE_IQ2_BN; break;
case GGML_FTYPE_MOSTLY_IQ4_NL: wtype = GGML_TYPE_IQ4_NL; break;
case GGML_FTYPE_MOSTLY_IQ4_XS: wtype = GGML_TYPE_IQ4_XS; break;
+ case GGML_FTYPE_MOSTLY_IQ4_K: wtype = GGML_TYPE_IQ4_K; break;
case GGML_FTYPE_MOSTLY_IQ3_S: wtype = GGML_TYPE_IQ3_S; break;
case GGML_FTYPE_MOSTLY_IQ2_S: wtype = GGML_TYPE_IQ2_S; break;
case GGML_FTYPE_MOSTLY_Q4_0_4_4: wtype = GGML_TYPE_Q4_0_4_4; break;
@@ -9577,6 +9591,7 @@ static void ggml_compute_forward_add(
case GGML_TYPE_IQ2_BN:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_XS:
+ case GGML_TYPE_IQ4_K:
case GGML_TYPE_IQ3_S:
case GGML_TYPE_IQ2_S:
case GGML_TYPE_Q4_0_4_4:
@@ -9957,6 +9972,7 @@ static void ggml_compute_forward_add1(
case GGML_TYPE_IQ2_BN:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_XS:
+ case GGML_TYPE_IQ4_K:
case GGML_TYPE_IQ3_S:
case GGML_TYPE_IQ2_S:
case GGML_TYPE_Q4_0_4_4:
@@ -10087,6 +10103,7 @@ static void ggml_compute_forward_acc(
case GGML_TYPE_IQ2_BN:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_XS:
+ case GGML_TYPE_IQ4_K:
case GGML_TYPE_IQ3_S:
case GGML_TYPE_IQ2_S:
case GGML_TYPE_Q4_0_4_4:
@@ -13006,6 +13023,7 @@ static void ggml_compute_forward_out_prod(
case GGML_TYPE_IQ2_BN:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_XS:
+ case GGML_TYPE_IQ4_K:
case GGML_TYPE_IQ3_S:
case GGML_TYPE_IQ2_S:
case GGML_TYPE_Q4_0_4_4:
@@ -13196,6 +13214,7 @@ static void ggml_compute_forward_set(
case GGML_TYPE_IQ2_BN:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_XS:
+ case GGML_TYPE_IQ4_K:
case GGML_TYPE_IQ3_S:
case GGML_TYPE_IQ2_S:
case GGML_TYPE_Q4_0_4_4:
@@ -13460,6 +13479,7 @@ static void ggml_compute_forward_get_rows(
case GGML_TYPE_IQ2_BN:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_XS:
+ case GGML_TYPE_IQ4_K:
case GGML_TYPE_IQ3_S:
case GGML_TYPE_IQ2_S:
case GGML_TYPE_Q4_0_4_4:
@@ -14051,6 +14071,7 @@ static void ggml_compute_forward_clamp(
case GGML_TYPE_IQ2_BN:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_XS:
+ case GGML_TYPE_IQ4_K:
case GGML_TYPE_IQ3_S:
case GGML_TYPE_IQ2_S:
case GGML_TYPE_Q8_K:
@@ -20786,6 +20807,7 @@ size_t ggml_quantize_chunk(
case GGML_TYPE_IQ2_BN: result = quantize_iq2_bn (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ4_NL: result = quantize_iq4_nl (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ4_XS: result = quantize_iq4_xs (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+ case GGML_TYPE_IQ4_K: result = quantize_iq4_k (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_Q4_0_4_4: result = quantize_q4_0_4x4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_Q4_0_4_8: result = quantize_q4_0_4x8(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_Q4_0_8_8: result = quantize_q4_0_8x8(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;