diff options
Diffstat (limited to 'ggml/src/ggml.c')
| -rw-r--r-- | ggml/src/ggml.c | 22 |
1 files changed, 22 insertions, 0 deletions
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c index 97b5bff7..2e6983df 100644 --- a/ggml/src/ggml.c +++ b/ggml/src/ggml.c @@ -1656,6 +1656,19 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = { .nrows = 1, .row_meta_size = 0, }, + [GGML_TYPE_IQ3_KS] = { + .type_name = "iq3_ks", + .blck_size = QK_K, + .type_size = sizeof(block_iq3_ks), + .is_quantized = true, + .to_float = (ggml_to_float_t) dequantize_row_iq3_ks, + .from_float = quantize_row_iq3_ks, + .from_float_ref = (ggml_from_float_t)quantize_row_iq3_ks_ref, + .vec_dot = vec_dot_iq3_ks_q8_k, + .vec_dot_type = GGML_TYPE_Q8_K, + .nrows = 1, + .row_meta_size = 2, + }, [GGML_TYPE_IQ4_K] = { .type_name = "iq4_k", .blck_size = QK_K, @@ -4578,6 +4591,7 @@ enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype) { case GGML_FTYPE_MOSTLY_IQ3_KT: wtype = GGML_TYPE_IQ3_KT; break; case GGML_FTYPE_MOSTLY_IQ4_KT: wtype = GGML_TYPE_IQ4_KT; break; case GGML_FTYPE_MOSTLY_IQ3_K: wtype = GGML_TYPE_IQ3_K; break; + case GGML_FTYPE_MOSTLY_IQ3_KS: wtype = GGML_TYPE_IQ3_KS; break; case GGML_FTYPE_MOSTLY_IQ4_K: wtype = GGML_TYPE_IQ4_K; break; case GGML_FTYPE_MOSTLY_IQ3_K_R4: wtype = GGML_TYPE_IQ3_K_R4; break; case GGML_FTYPE_MOSTLY_IQ4_K_R4: wtype = GGML_TYPE_IQ4_K_R4; break; @@ -11347,6 +11361,7 @@ static void ggml_compute_forward_add( case GGML_TYPE_IQ3_KT: case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: + case GGML_TYPE_IQ3_KS: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: case GGML_TYPE_IQ4_K_R4: @@ -11824,6 +11839,7 @@ static void ggml_compute_forward_add1( case GGML_TYPE_IQ3_KT: case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: + case GGML_TYPE_IQ3_KS: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: case GGML_TYPE_IQ4_K_R4: @@ -11998,6 +12014,7 @@ static void ggml_compute_forward_acc( case GGML_TYPE_IQ3_KT: case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: + case GGML_TYPE_IQ3_KS: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: case GGML_TYPE_IQ4_K_R4: @@ -15499,6 +15516,7 @@ static void ggml_compute_forward_out_prod( case GGML_TYPE_IQ3_KT: case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: + case GGML_TYPE_IQ3_KS: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: case GGML_TYPE_IQ4_K_R4: @@ -15913,6 +15931,7 @@ static void ggml_compute_forward_set( case GGML_TYPE_IQ3_KT: case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: + case GGML_TYPE_IQ3_KS: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: case GGML_TYPE_IQ4_K_R4: @@ -16233,6 +16252,7 @@ static void ggml_compute_forward_get_rows( case GGML_TYPE_IQ3_KT: case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: + case GGML_TYPE_IQ3_KS: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: case GGML_TYPE_IQ4_K_R4: @@ -16870,6 +16890,7 @@ static void ggml_compute_forward_clamp( case GGML_TYPE_IQ3_KT: case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: + case GGML_TYPE_IQ3_KS: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: case GGML_TYPE_IQ4_K_R4: @@ -23943,6 +23964,7 @@ size_t ggml_quantize_chunk( case GGML_TYPE_IQ3_KT: result = quantize_iq3_kt (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; case GGML_TYPE_IQ4_KT: result = quantize_iq4_kt (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; case GGML_TYPE_IQ3_K: result = quantize_iq3_k (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; + case GGML_TYPE_IQ3_KS: result = quantize_iq3_ks (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_IQ3_K_R4:result = quantize_iq3_k_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; case GGML_TYPE_IQ4_K_R4:result = quantize_iq4_k_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; |