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.c90
1 files changed, 45 insertions, 45 deletions
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index 4199a282..68525906 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -1606,28 +1606,28 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
.nrows = 1,
.row_meta_size = 0,
},
- [GGML_TYPE_IQ4_XS_R4] = {
- .type_name = "iq4_xs_r4",
+ [GGML_TYPE_IQ4_XS_R8] = {
+ .type_name = "iq4_xs_r8",
.blck_size = QK_K,
.type_size = sizeof(block_iq4_xs),
.is_quantized = true,
- .to_float = (ggml_to_float_t) dequantize_row_iq4_xs_r4,
- .from_float = quantize_row_iq4_xs_r4,
- .from_float_ref = (ggml_from_float_t)quantize_row_iq4_xs_r4_ref,
- .vec_dot = vec_dot_iq4_xs_r4_q8_k,
+ .to_float = (ggml_to_float_t) dequantize_row_iq4_xs_r8,
+ .from_float = quantize_row_iq4_xs_r8,
+ .from_float_ref = (ggml_from_float_t)quantize_row_iq4_xs_r8_ref,
+ .vec_dot = vec_dot_iq4_xs_r8_q8_k,
.vec_dot_type = GGML_TYPE_Q8_K32,
.nrows = 1,
.row_meta_size = 0,
},
- [GGML_TYPE_Q4_0_R4] = {
- .type_name = "q4_0_r4",
+ [GGML_TYPE_Q4_0_R8] = {
+ .type_name = "q4_0_r8",
.blck_size = QK4_NL,
.type_size = sizeof(block_iq4_nl),
.is_quantized = true,
- .to_float = (ggml_to_float_t) dequantize_row_q4_0_r4,
- .from_float = quantize_row_q4_0_r4,
- .from_float_ref = (ggml_from_float_t)quantize_row_q4_0_r4_ref,
- .vec_dot = vec_dot_q4_0_r4_q8_0,
+ .to_float = (ggml_to_float_t) dequantize_row_q4_0_r8,
+ .from_float = quantize_row_q4_0_r8,
+ .from_float_ref = (ggml_from_float_t)quantize_row_q4_0_r8_ref,
+ .vec_dot = vec_dot_q4_0_r8_q8_0,
#if GGML_USE_IQK_MULMAT
#if defined __AVX2__
.vec_dot_type = GGML_TYPE_Q8_1_X4,
@@ -1640,15 +1640,15 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
.nrows = 1,
.row_meta_size = 0,
},
- [GGML_TYPE_Q8_0_R4] = {
- .type_name = "q8_0_r4",
+ [GGML_TYPE_Q8_0_R8] = {
+ .type_name = "q8_0_r8",
.blck_size = QK8_0,
.type_size = sizeof(block_q8_0),
.is_quantized = true,
- .to_float = (ggml_to_float_t) dequantize_row_q8_0_r4,
- .from_float = quantize_row_q8_0_r4,
- .from_float_ref = (ggml_from_float_t)quantize_row_q8_0_r4_ref,
- .vec_dot = vec_dot_q8_0_r4_q8_0,
+ .to_float = (ggml_to_float_t) dequantize_row_q8_0_r8,
+ .from_float = quantize_row_q8_0_r8,
+ .from_float_ref = (ggml_from_float_t)quantize_row_q8_0_r8_ref,
+ .vec_dot = vec_dot_q8_0_r8_q8_0,
#if GGML_USE_IQK_MULMAT
#if defined __AVX2__
.vec_dot_type = GGML_TYPE_Q8_1_X4,
@@ -4390,11 +4390,11 @@ enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype) {
case GGML_FTYPE_MOSTLY_IQ2_BN_R4: wtype = GGML_TYPE_IQ2_BN_R4;break;
case GGML_FTYPE_MOSTLY_IQ4_NL: wtype = GGML_TYPE_IQ4_NL; break;
case GGML_FTYPE_MOSTLY_IQ4_NL_R4: wtype = GGML_TYPE_IQ4_NL_R4;break;
- case GGML_FTYPE_MOSTLY_IQ4_XS_R4: wtype = GGML_TYPE_IQ4_XS_R4;break;
- case GGML_FTYPE_MOSTLY_Q4_0_R4: wtype = GGML_TYPE_Q4_0_R4; break;
+ case GGML_FTYPE_MOSTLY_IQ4_XS_R8: wtype = GGML_TYPE_IQ4_XS_R8;break;
+ case GGML_FTYPE_MOSTLY_Q4_0_R8: wtype = GGML_TYPE_Q4_0_R8; break;
case GGML_FTYPE_MOSTLY_Q5_0_R4: wtype = GGML_TYPE_Q5_0_R4; break;
case GGML_FTYPE_MOSTLY_Q6_0_R4: wtype = GGML_TYPE_Q6_0_R4; break;
- case GGML_FTYPE_MOSTLY_Q8_0_R4: wtype = GGML_TYPE_Q8_0_R4; break;
+ case GGML_FTYPE_MOSTLY_Q8_0_R8: wtype = GGML_TYPE_Q8_0_R8; break;
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;
@@ -10938,12 +10938,12 @@ static void ggml_compute_forward_add(
case GGML_TYPE_IQ2_BN_R4:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_NL_R4:
- case GGML_TYPE_IQ4_XS_R4:
- case GGML_TYPE_Q4_0_R4:
+ case GGML_TYPE_IQ4_XS_R8:
+ case GGML_TYPE_Q4_0_R8:
case GGML_TYPE_Q5_0_R4:
case GGML_TYPE_Q6_0_R4:
case GGML_TYPE_I2_S:
- case GGML_TYPE_Q8_0_R4:
+ case GGML_TYPE_Q8_0_R8:
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
@@ -11408,12 +11408,12 @@ static void ggml_compute_forward_add1(
case GGML_TYPE_IQ2_BN_R4:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_NL_R4:
- case GGML_TYPE_IQ4_XS_R4:
- case GGML_TYPE_Q4_0_R4:
+ case GGML_TYPE_IQ4_XS_R8:
+ case GGML_TYPE_Q4_0_R8:
case GGML_TYPE_Q5_0_R4:
case GGML_TYPE_Q6_0_R4:
case GGML_TYPE_I2_S:
- case GGML_TYPE_Q8_0_R4:
+ case GGML_TYPE_Q8_0_R8:
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
@@ -11575,12 +11575,12 @@ static void ggml_compute_forward_acc(
case GGML_TYPE_IQ2_BN_R4:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_NL_R4:
- case GGML_TYPE_IQ4_XS_R4:
- case GGML_TYPE_Q4_0_R4:
+ case GGML_TYPE_IQ4_XS_R8:
+ case GGML_TYPE_Q4_0_R8:
case GGML_TYPE_Q5_0_R4:
case GGML_TYPE_Q6_0_R4:
case GGML_TYPE_I2_S:
- case GGML_TYPE_Q8_0_R4:
+ case GGML_TYPE_Q8_0_R8:
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
@@ -14815,12 +14815,12 @@ static void ggml_compute_forward_out_prod(
case GGML_TYPE_IQ2_BN_R4:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_NL_R4:
- case GGML_TYPE_IQ4_XS_R4:
- case GGML_TYPE_Q4_0_R4:
+ case GGML_TYPE_IQ4_XS_R8:
+ case GGML_TYPE_Q4_0_R8:
case GGML_TYPE_Q5_0_R4:
case GGML_TYPE_Q6_0_R4:
case GGML_TYPE_I2_S:
- case GGML_TYPE_Q8_0_R4:
+ case GGML_TYPE_Q8_0_R8:
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
@@ -15222,12 +15222,12 @@ static void ggml_compute_forward_set(
case GGML_TYPE_IQ2_BN_R4:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_NL_R4:
- case GGML_TYPE_IQ4_XS_R4:
- case GGML_TYPE_Q4_0_R4:
+ case GGML_TYPE_IQ4_XS_R8:
+ case GGML_TYPE_Q4_0_R8:
case GGML_TYPE_Q5_0_R4:
case GGML_TYPE_Q6_0_R4:
case GGML_TYPE_I2_S:
- case GGML_TYPE_Q8_0_R4:
+ case GGML_TYPE_Q8_0_R8:
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
@@ -15523,12 +15523,12 @@ static void ggml_compute_forward_get_rows(
case GGML_TYPE_IQ2_BN_R4:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_NL_R4:
- case GGML_TYPE_IQ4_XS_R4:
- case GGML_TYPE_Q4_0_R4:
+ case GGML_TYPE_IQ4_XS_R8:
+ case GGML_TYPE_Q4_0_R8:
case GGML_TYPE_Q5_0_R4:
case GGML_TYPE_Q6_0_R4:
case GGML_TYPE_I2_S:
- case GGML_TYPE_Q8_0_R4:
+ case GGML_TYPE_Q8_0_R8:
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
@@ -16153,12 +16153,12 @@ static void ggml_compute_forward_clamp(
case GGML_TYPE_IQ2_BN_R4:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_NL_R4:
- case GGML_TYPE_IQ4_XS_R4:
- case GGML_TYPE_Q4_0_R4:
+ case GGML_TYPE_IQ4_XS_R8:
+ case GGML_TYPE_Q4_0_R8:
case GGML_TYPE_Q5_0_R4:
case GGML_TYPE_Q6_0_R4:
case GGML_TYPE_I2_S:
- case GGML_TYPE_Q8_0_R4:
+ case GGML_TYPE_Q8_0_R8:
case GGML_TYPE_IQ4_XS:
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
@@ -23028,11 +23028,11 @@ size_t ggml_quantize_chunk(
case GGML_TYPE_IQ2_BN_R4:result = quantize_iq2_bn_r4(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_NL_R4: result = quantize_iq4_nl_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
- case GGML_TYPE_IQ4_XS_R4: result = quantize_iq4_xs_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
- case GGML_TYPE_Q4_0_R4: result = quantize_q4_0_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+ case GGML_TYPE_IQ4_XS_R8: result = quantize_iq4_xs_r8(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+ case GGML_TYPE_Q4_0_R8: result = quantize_q4_0_r8(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_Q5_0_R4: result = quantize_q5_0_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
case GGML_TYPE_Q6_0_R4: result = quantize_q6_0_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
- case GGML_TYPE_Q8_0_R4: result = quantize_q8_0_r4(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+ case GGML_TYPE_Q8_0_R8: result = quantize_q8_0_r8(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_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;
generated by cgit v1.2.3 (git 2.25.1) at 2025-08-17 14:15:09 +0000