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.c26
1 files changed, 26 insertions, 0 deletions
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index bc103ab7..7cbc0056 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -1343,6 +1343,23 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
.nrows = 1,
.row_meta_size = 4,
},
+ [GGML_TYPE_IQ5_KS_R4] = {
+ .type_name = "iq5_ks_r4",
+ .blck_size = QK_K,
+ .type_size = sizeof(block_iq5_ks),
+ .is_quantized = true,
+ .to_float = (ggml_to_float_t) dequantize_row_iq5_ks_r4,
+ .from_float = quantize_row_iq5_ks_r4,
+ .from_float_ref = (ggml_from_float_t)quantize_row_iq5_ks_r4_ref,
+ .vec_dot = vec_dot_iq5_ks_r4_q8_k,
+#if defined __AVX2__
+ .vec_dot_type = GGML_TYPE_Q8_K32,
+#else
+ .vec_dot_type = GGML_TYPE_Q8_K,
+#endif
+ .nrows = 1,
+ .row_meta_size = 4,
+ },
[GGML_TYPE_IQ4_KSS] = {
.type_name = "iq4_kss",
.blck_size = QK_K,
@@ -4478,6 +4495,7 @@ enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype) {
case GGML_FTYPE_MOSTLY_IQ4_XS: wtype = GGML_TYPE_IQ4_XS; break;
case GGML_FTYPE_MOSTLY_IQ4_KS: wtype = GGML_TYPE_IQ4_KS; break;
case GGML_FTYPE_MOSTLY_IQ4_KS_R4: wtype = GGML_TYPE_IQ4_KS_R4;break;
+ case GGML_FTYPE_MOSTLY_IQ5_KS_R4: wtype = GGML_TYPE_IQ5_KS_R4;break;
case GGML_FTYPE_MOSTLY_IQ4_KSS: wtype = GGML_TYPE_IQ4_KSS; break;
case GGML_FTYPE_MOSTLY_IQ5_KS: wtype = GGML_TYPE_IQ5_KS; break;
case GGML_FTYPE_MOSTLY_IQ2_K: wtype = GGML_TYPE_IQ2_K; break;
@@ -11242,6 +11260,7 @@ static void ggml_compute_forward_add(
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
+ case GGML_TYPE_IQ5_KS_R4:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ5_KS:
case GGML_TYPE_IQ2_K:
@@ -11715,6 +11734,7 @@ static void ggml_compute_forward_add1(
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
+ case GGML_TYPE_IQ5_KS_R4:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ5_KS:
case GGML_TYPE_IQ2_K:
@@ -11885,6 +11905,7 @@ static void ggml_compute_forward_acc(
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
+ case GGML_TYPE_IQ5_KS_R4:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ5_KS:
case GGML_TYPE_IQ2_K:
@@ -15382,6 +15403,7 @@ static void ggml_compute_forward_out_prod(
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
+ case GGML_TYPE_IQ5_KS_R4:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ5_KS:
case GGML_TYPE_IQ2_K:
@@ -15792,6 +15814,7 @@ static void ggml_compute_forward_set(
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
+ case GGML_TYPE_IQ5_KS_R4:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ5_KS:
case GGML_TYPE_IQ2_K:
@@ -16108,6 +16131,7 @@ static void ggml_compute_forward_get_rows(
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
+ case GGML_TYPE_IQ5_KS_R4:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ5_KS:
case GGML_TYPE_IQ2_K:
@@ -16741,6 +16765,7 @@ static void ggml_compute_forward_clamp(
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
+ case GGML_TYPE_IQ5_KS_R4:
case GGML_TYPE_IQ4_KSS:
case GGML_TYPE_IQ5_KS:
case GGML_TYPE_IQ2_K:
@@ -23810,6 +23835,7 @@ size_t ggml_quantize_chunk(
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_KS: result = quantize_iq4_ks (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ4_KS_R4:result = quantize_iq4_ks_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+ case GGML_TYPE_IQ5_KS_R4:result = quantize_iq5_ks_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ4_KSS: result = quantize_iq4_kss(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ5_KS: result = quantize_iq5_ks (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_IQ2_K: result = quantize_iq2_k (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;