diff options
Diffstat (limited to 'ggml.c')
-rw-r--r-- | ggml.c | 28 |
1 files changed, 26 insertions, 2 deletions
@@ -794,6 +794,18 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = { .vec_dot_type = GGML_TYPE_Q8_K, .nrows = 1, }, + [GGML_TYPE_IQ1_M] = { + .type_name = "iq1_m", + .blck_size = QK_K, + .type_size = sizeof(block_iq1_m), + .is_quantized = true, + .to_float = (ggml_to_float_t) dequantize_row_iq1_m, + .from_float = NULL, + .from_float_reference = NULL, + .vec_dot = ggml_vec_dot_iq1_m_q8_K, + .vec_dot_type = GGML_TYPE_Q8_K, + .nrows = 1, + }, [GGML_TYPE_IQ4_NL] = { .type_name = "iq4_nl", .blck_size = QK4_NL, @@ -2539,6 +2551,7 @@ enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype) { case GGML_FTYPE_MOSTLY_IQ2_XS: wtype = GGML_TYPE_IQ2_XS; break; case GGML_FTYPE_MOSTLY_IQ3_XXS: wtype = GGML_TYPE_IQ3_XXS; break; case GGML_FTYPE_MOSTLY_IQ1_S: wtype = GGML_TYPE_IQ1_S; break; + case GGML_FTYPE_MOSTLY_IQ1_M: wtype = GGML_TYPE_IQ1_M; 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_IQ3_S: wtype = GGML_TYPE_IQ3_S; break; @@ -8135,6 +8148,7 @@ static void ggml_compute_forward_add( case GGML_TYPE_IQ2_XS: case GGML_TYPE_IQ3_XXS: case GGML_TYPE_IQ1_S: + case GGML_TYPE_IQ1_M: case GGML_TYPE_IQ4_NL: case GGML_TYPE_IQ4_XS: case GGML_TYPE_IQ3_S: @@ -8417,6 +8431,7 @@ static void ggml_compute_forward_add1( case GGML_TYPE_IQ2_XS: case GGML_TYPE_IQ3_XXS: case GGML_TYPE_IQ1_S: + case GGML_TYPE_IQ1_M: case GGML_TYPE_IQ4_NL: case GGML_TYPE_IQ4_XS: case GGML_TYPE_IQ3_S: @@ -8544,6 +8559,7 @@ static void ggml_compute_forward_acc( case GGML_TYPE_IQ2_XS: case GGML_TYPE_IQ3_XXS: case GGML_TYPE_IQ1_S: + case GGML_TYPE_IQ1_M: case GGML_TYPE_IQ4_NL: case GGML_TYPE_IQ4_XS: case GGML_TYPE_IQ3_S: @@ -11447,6 +11463,7 @@ static void ggml_compute_forward_out_prod( case GGML_TYPE_IQ2_XS: case GGML_TYPE_IQ3_XXS: case GGML_TYPE_IQ1_S: + case GGML_TYPE_IQ1_M: case GGML_TYPE_IQ4_NL: case GGML_TYPE_IQ4_XS: case GGML_TYPE_IQ3_S: @@ -11638,6 +11655,7 @@ static void ggml_compute_forward_set( case GGML_TYPE_IQ2_XS: case GGML_TYPE_IQ3_XXS: case GGML_TYPE_IQ1_S: + case GGML_TYPE_IQ1_M: case GGML_TYPE_IQ4_NL: case GGML_TYPE_IQ4_XS: case GGML_TYPE_IQ3_S: @@ -11861,6 +11879,7 @@ static void ggml_compute_forward_get_rows( case GGML_TYPE_IQ2_XS: case GGML_TYPE_IQ3_XXS: case GGML_TYPE_IQ1_S: + case GGML_TYPE_IQ1_M: case GGML_TYPE_IQ4_NL: case GGML_TYPE_IQ4_XS: case GGML_TYPE_IQ3_S: @@ -12564,6 +12583,7 @@ static void ggml_compute_forward_alibi( case GGML_TYPE_IQ2_XS: case GGML_TYPE_IQ3_XXS: case GGML_TYPE_IQ1_S: + case GGML_TYPE_IQ1_M: case GGML_TYPE_IQ4_NL: case GGML_TYPE_IQ4_XS: case GGML_TYPE_IQ3_S: @@ -12652,6 +12672,7 @@ static void ggml_compute_forward_clamp( case GGML_TYPE_IQ2_XS: case GGML_TYPE_IQ3_XXS: case GGML_TYPE_IQ1_S: + case GGML_TYPE_IQ1_M: case GGML_TYPE_IQ4_NL: case GGML_TYPE_IQ4_XS: case GGML_TYPE_IQ3_S: @@ -20306,7 +20327,8 @@ void ggml_quantize_init(enum ggml_type type) { case GGML_TYPE_IQ2_XXS: case GGML_TYPE_IQ2_XS: case GGML_TYPE_IQ2_S: - case GGML_TYPE_IQ1_S: iq2xs_init_impl(type); break; + case GGML_TYPE_IQ1_S: + case GGML_TYPE_IQ1_M: iq2xs_init_impl(type); break; case GGML_TYPE_IQ3_XXS: iq3xs_init_impl(256); break; case GGML_TYPE_IQ3_S: iq3xs_init_impl(512); break; default: // nothing @@ -20331,7 +20353,8 @@ bool ggml_quantize_requires_imatrix(enum ggml_type type) { return type == GGML_TYPE_IQ2_XXS || type == GGML_TYPE_IQ2_XS || - type == GGML_TYPE_IQ1_S; + type == GGML_TYPE_IQ1_S;// || + //type == GGML_TYPE_IQ1_M; } size_t ggml_quantize_chunk( @@ -20375,6 +20398,7 @@ size_t ggml_quantize_chunk( case GGML_TYPE_IQ3_S: result = quantize_iq3_s (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; case GGML_TYPE_IQ2_S: result = quantize_iq2_s (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; case GGML_TYPE_IQ1_S: result = quantize_iq1_s (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; + case GGML_TYPE_IQ1_M: result = quantize_iq1_m (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; #if QK_K == 64 case GGML_TYPE_IQ4_XS: result = quantize_iq4_nl (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break; |