diff options
Diffstat (limited to 'ggml/src/ggml.c')
| -rw-r--r-- | ggml/src/ggml.c | 66 |
1 files changed, 66 insertions, 0 deletions
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c index 7cbc0056..d8025a5a 100644 --- a/ggml/src/ggml.c +++ b/ggml/src/ggml.c @@ -1574,6 +1574,45 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = { .nrows = 1, .row_meta_size = 2, }, + [GGML_TYPE_IQ2_KT] = { + .type_name = "iq2_kt", + .blck_size = QK_K, + .type_size = sizeof(block_iq2_kt), + .is_quantized = true, + .to_float = (ggml_to_float_t) dequantize_row_iq2_kt, + .from_float = quantize_row_iq2_kt, + .from_float_ref = (ggml_from_float_t)quantize_row_iq2_kt_ref, + .vec_dot = vec_dot_iq2_kt_q8_k, + .vec_dot_type = GGML_TYPE_Q8_K, + .nrows = 1, + .row_meta_size = 4, + }, + [GGML_TYPE_IQ3_KT] = { + .type_name = "iq3_kt", + .blck_size = QK_K, + .type_size = sizeof(block_iq3_kt), + .is_quantized = true, + .to_float = (ggml_to_float_t) dequantize_row_iq3_kt, + .from_float = quantize_row_iq3_kt, + .from_float_ref = (ggml_from_float_t)quantize_row_iq3_kt_ref, + .vec_dot = vec_dot_iq3_kt_q8_k, + .vec_dot_type = GGML_TYPE_Q8_K, + .nrows = 1, + .row_meta_size = 4, + }, + [GGML_TYPE_IQ4_KT] = { + .type_name = "iq4_kt", + .blck_size = QK_K, + .type_size = sizeof(block_iq4_kt), + .is_quantized = true, + .to_float = (ggml_to_float_t) dequantize_row_iq4_kt, + .from_float = quantize_row_iq4_kt, + .from_float_ref = (ggml_from_float_t)quantize_row_iq4_kt_ref, + .vec_dot = vec_dot_iq4_kt_q8_k, + .vec_dot_type = GGML_TYPE_Q8_K, + .nrows = 1, + .row_meta_size = 8, + }, [GGML_TYPE_IQ3_K] = { .type_name = "iq3_k", .blck_size = QK_K, @@ -4501,6 +4540,9 @@ enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype) { case GGML_FTYPE_MOSTLY_IQ2_K: wtype = GGML_TYPE_IQ2_K; break; case GGML_FTYPE_MOSTLY_IQ2_K_R4: wtype = GGML_TYPE_IQ2_K_R4; break; case GGML_FTYPE_MOSTLY_IQ2_KS: wtype = GGML_TYPE_IQ2_KS; break; + case GGML_FTYPE_MOSTLY_IQ2_KT: wtype = GGML_TYPE_IQ2_KT; break; + 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_IQ4_K: wtype = GGML_TYPE_IQ4_K; break; case GGML_FTYPE_MOSTLY_IQ3_K_R4: wtype = GGML_TYPE_IQ3_K_R4; break; @@ -11266,6 +11308,9 @@ static void ggml_compute_forward_add( case GGML_TYPE_IQ2_K: case GGML_TYPE_IQ2_K_R4: case GGML_TYPE_IQ2_KS: + case GGML_TYPE_IQ2_KT: + case GGML_TYPE_IQ3_KT: + case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: @@ -11740,6 +11785,9 @@ static void ggml_compute_forward_add1( case GGML_TYPE_IQ2_K: case GGML_TYPE_IQ2_K_R4: case GGML_TYPE_IQ2_KS: + case GGML_TYPE_IQ2_KT: + case GGML_TYPE_IQ3_KT: + case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: @@ -11911,6 +11959,9 @@ static void ggml_compute_forward_acc( case GGML_TYPE_IQ2_K: case GGML_TYPE_IQ2_K_R4: case GGML_TYPE_IQ2_KS: + case GGML_TYPE_IQ2_KT: + case GGML_TYPE_IQ3_KT: + case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: @@ -15409,6 +15460,9 @@ static void ggml_compute_forward_out_prod( case GGML_TYPE_IQ2_K: case GGML_TYPE_IQ2_K_R4: case GGML_TYPE_IQ2_KS: + case GGML_TYPE_IQ2_KT: + case GGML_TYPE_IQ3_KT: + case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: @@ -15820,6 +15874,9 @@ static void ggml_compute_forward_set( case GGML_TYPE_IQ2_K: case GGML_TYPE_IQ2_K_R4: case GGML_TYPE_IQ2_KS: + case GGML_TYPE_IQ2_KT: + case GGML_TYPE_IQ3_KT: + case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: @@ -16137,6 +16194,9 @@ static void ggml_compute_forward_get_rows( case GGML_TYPE_IQ2_K: case GGML_TYPE_IQ2_K_R4: case GGML_TYPE_IQ2_KS: + case GGML_TYPE_IQ2_KT: + case GGML_TYPE_IQ3_KT: + case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: @@ -16771,6 +16831,9 @@ static void ggml_compute_forward_clamp( case GGML_TYPE_IQ2_K: case GGML_TYPE_IQ2_K_R4: case GGML_TYPE_IQ2_KS: + case GGML_TYPE_IQ2_KT: + case GGML_TYPE_IQ3_KT: + case GGML_TYPE_IQ4_KT: case GGML_TYPE_IQ3_K: case GGML_TYPE_IQ4_K: case GGML_TYPE_IQ3_K_R4: @@ -23841,6 +23904,9 @@ size_t ggml_quantize_chunk( case GGML_TYPE_IQ2_K: result = quantize_iq2_k (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; case GGML_TYPE_IQ2_K_R4:result = quantize_iq2_k_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; case GGML_TYPE_IQ2_KS: result = quantize_iq2_ks (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; + case GGML_TYPE_IQ2_KT: result = quantize_iq2_kt (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; + 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_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; |