ggml : minor naming changes (#8433)
* ggml : minor naming changes ggml-ci * ggml : use PRId64 [no ci] * ggml : revert FA K/Q names
This commit is contained in:
Georgi Gerganov
GitHub
parent
b549a1bbef
commit
370b1f7e7a
10 changed files with 192 additions and 189 deletions
111
ggml/src/ggml.c
111
ggml/src/ggml.c
|
|
@ -592,7 +592,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = false,
|
||||
.to_float = (ggml_to_float_t) ggml_fp16_to_fp32_row,
|
||||
.from_float = (ggml_from_float_t) ggml_fp32_to_fp16_row,
|
||||
.from_float_reference = (ggml_from_float_t) ggml_fp32_to_fp16_row,
|
||||
.from_float_ref = (ggml_from_float_t) ggml_fp32_to_fp16_row,
|
||||
.vec_dot = (ggml_vec_dot_t) ggml_vec_dot_f16,
|
||||
.vec_dot_type = GGML_TYPE_F16,
|
||||
.nrows = 1,
|
||||
|
|
@ -604,7 +604,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_q4_0,
|
||||
.from_float = quantize_row_q4_0,
|
||||
.from_float_reference = (ggml_from_float_t) quantize_row_q4_0_reference,
|
||||
.from_float_ref = (ggml_from_float_t) quantize_row_q4_0_ref,
|
||||
.vec_dot = ggml_vec_dot_q4_0_q8_0,
|
||||
.vec_dot_type = GGML_TYPE_Q8_0,
|
||||
#if defined (__ARM_FEATURE_MATMUL_INT8)
|
||||
|
|
@ -620,7 +620,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_q4_1,
|
||||
.from_float = quantize_row_q4_1,
|
||||
.from_float_reference = (ggml_from_float_t) quantize_row_q4_1_reference,
|
||||
.from_float_ref = (ggml_from_float_t) quantize_row_q4_1_ref,
|
||||
.vec_dot = ggml_vec_dot_q4_1_q8_1,
|
||||
.vec_dot_type = GGML_TYPE_Q8_1,
|
||||
#if defined (__ARM_FEATURE_MATMUL_INT8)
|
||||
|
|
@ -636,7 +636,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = false,
|
||||
.to_float = NULL,
|
||||
.from_float = NULL,
|
||||
.from_float_reference = NULL,
|
||||
.from_float_ref = NULL,
|
||||
.vec_dot = NULL,
|
||||
.vec_dot_type = GGML_TYPE_COUNT,
|
||||
.nrows = 1,
|
||||
|
|
@ -648,7 +648,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = false,
|
||||
.to_float = NULL,
|
||||
.from_float = NULL,
|
||||
.from_float_reference = NULL,
|
||||
.from_float_ref = NULL,
|
||||
.vec_dot = NULL,
|
||||
.vec_dot_type = GGML_TYPE_COUNT,
|
||||
.nrows = 1,
|
||||
|
|
@ -660,7 +660,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_q5_0,
|
||||
.from_float = quantize_row_q5_0,
|
||||
.from_float_reference = (ggml_from_float_t) quantize_row_q5_0_reference,
|
||||
.from_float_ref = (ggml_from_float_t) quantize_row_q5_0_ref,
|
||||
.vec_dot = ggml_vec_dot_q5_0_q8_0,
|
||||
.vec_dot_type = GGML_TYPE_Q8_0,
|
||||
.nrows = 1,
|
||||
|
|
@ -672,7 +672,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_q5_1,
|
||||
.from_float = quantize_row_q5_1,
|
||||
.from_float_reference = (ggml_from_float_t) quantize_row_q5_1_reference,
|
||||
.from_float_ref = (ggml_from_float_t) quantize_row_q5_1_ref,
|
||||
.vec_dot = ggml_vec_dot_q5_1_q8_1,
|
||||
.vec_dot_type = GGML_TYPE_Q8_1,
|
||||
.nrows = 1,
|
||||
|
|
@ -684,7 +684,8 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_q8_0,
|
||||
.from_float = quantize_row_q8_0,
|
||||
.from_float_reference = (ggml_from_float_t) quantize_row_q8_0_reference,
|
||||
.from_float_ref = (ggml_from_float_t) quantize_row_q8_0_ref,
|
||||
.from_float_to_mat = quantize_mat_q8_0,
|
||||
.vec_dot = ggml_vec_dot_q8_0_q8_0,
|
||||
.vec_dot_type = GGML_TYPE_Q8_0,
|
||||
#if defined (__ARM_FEATURE_MATMUL_INT8)
|
||||
|
|
@ -692,7 +693,6 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
#else
|
||||
.nrows = 1,
|
||||
#endif
|
||||
.from_float_to_mat = quantize_mat_q8_0,
|
||||
},
|
||||
[GGML_TYPE_Q8_1] = {
|
||||
.type_name = "q8_1",
|
||||
|
|
@ -700,7 +700,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.type_size = sizeof(block_q8_1),
|
||||
.is_quantized = true,
|
||||
.from_float = quantize_row_q8_1,
|
||||
.from_float_reference = (ggml_from_float_t) quantize_row_q8_1_reference,
|
||||
.from_float_ref = (ggml_from_float_t) quantize_row_q8_1_ref,
|
||||
.vec_dot_type = GGML_TYPE_Q8_1,
|
||||
.nrows = 1,
|
||||
},
|
||||
|
|
@ -711,7 +711,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_q2_K,
|
||||
.from_float = quantize_row_q2_K,
|
||||
.from_float_reference = (ggml_from_float_t) quantize_row_q2_K_reference,
|
||||
.from_float_ref = (ggml_from_float_t) quantize_row_q2_K_ref,
|
||||
.vec_dot = ggml_vec_dot_q2_K_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -723,7 +723,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_q3_K,
|
||||
.from_float = quantize_row_q3_K,
|
||||
.from_float_reference = (ggml_from_float_t) quantize_row_q3_K_reference,
|
||||
.from_float_ref = (ggml_from_float_t) quantize_row_q3_K_ref,
|
||||
.vec_dot = ggml_vec_dot_q3_K_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -735,7 +735,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_q4_K,
|
||||
.from_float = quantize_row_q4_K,
|
||||
.from_float_reference = (ggml_from_float_t) quantize_row_q4_K_reference,
|
||||
.from_float_ref = (ggml_from_float_t) quantize_row_q4_K_ref,
|
||||
.vec_dot = ggml_vec_dot_q4_K_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -747,7 +747,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_q5_K,
|
||||
.from_float = quantize_row_q5_K,
|
||||
.from_float_reference = (ggml_from_float_t) quantize_row_q5_K_reference,
|
||||
.from_float_ref = (ggml_from_float_t) quantize_row_q5_K_ref,
|
||||
.vec_dot = ggml_vec_dot_q5_K_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -759,7 +759,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_q6_K,
|
||||
.from_float = quantize_row_q6_K,
|
||||
.from_float_reference = (ggml_from_float_t) quantize_row_q6_K_reference,
|
||||
.from_float_ref = (ggml_from_float_t) quantize_row_q6_K_ref,
|
||||
.vec_dot = ggml_vec_dot_q6_K_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -771,7 +771,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_iq2_xxs,
|
||||
.from_float = NULL,
|
||||
.from_float_reference = NULL,
|
||||
.from_float_ref = NULL,
|
||||
.vec_dot = ggml_vec_dot_iq2_xxs_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -783,7 +783,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_iq2_xs,
|
||||
.from_float = NULL,
|
||||
.from_float_reference = NULL,
|
||||
.from_float_ref = NULL,
|
||||
.vec_dot = ggml_vec_dot_iq2_xs_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -795,7 +795,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_iq3_xxs,
|
||||
.from_float = quantize_row_iq3_xxs,
|
||||
.from_float_reference = (ggml_from_float_t)quantize_row_iq3_xxs_reference,
|
||||
.from_float_ref = (ggml_from_float_t)quantize_row_iq3_xxs_ref,
|
||||
.vec_dot = ggml_vec_dot_iq3_xxs_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -807,7 +807,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_iq3_s,
|
||||
.from_float = quantize_row_iq3_s,
|
||||
.from_float_reference = (ggml_from_float_t)quantize_row_iq3_s_reference,
|
||||
.from_float_ref = (ggml_from_float_t)quantize_row_iq3_s_ref,
|
||||
.vec_dot = ggml_vec_dot_iq3_s_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -819,7 +819,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_iq2_s,
|
||||
.from_float = quantize_row_iq2_s,
|
||||
.from_float_reference = (ggml_from_float_t)quantize_row_iq2_s_reference,
|
||||
.from_float_ref = (ggml_from_float_t)quantize_row_iq2_s_ref,
|
||||
.vec_dot = ggml_vec_dot_iq2_s_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -831,7 +831,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_iq1_s,
|
||||
.from_float = NULL,
|
||||
.from_float_reference = NULL,
|
||||
.from_float_ref = NULL,
|
||||
.vec_dot = ggml_vec_dot_iq1_s_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -843,7 +843,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_iq1_m,
|
||||
.from_float = NULL,
|
||||
.from_float_reference = NULL,
|
||||
.from_float_ref = NULL,
|
||||
.vec_dot = ggml_vec_dot_iq1_m_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -855,7 +855,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_iq4_nl,
|
||||
.from_float = quantize_row_iq4_nl,
|
||||
.from_float_reference = (ggml_from_float_t)quantize_row_iq4_nl_reference,
|
||||
.from_float_ref = (ggml_from_float_t)quantize_row_iq4_nl_ref,
|
||||
.vec_dot = ggml_vec_dot_iq4_nl_q8_0,
|
||||
.vec_dot_type = GGML_TYPE_Q8_0,
|
||||
.nrows = 1,
|
||||
|
|
@ -867,7 +867,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = true,
|
||||
.to_float = (ggml_to_float_t) dequantize_row_iq4_xs,
|
||||
.from_float = quantize_row_iq4_xs,
|
||||
.from_float_reference = (ggml_from_float_t)quantize_row_iq4_xs_reference,
|
||||
.from_float_ref = (ggml_from_float_t)quantize_row_iq4_xs_ref,
|
||||
.vec_dot = ggml_vec_dot_iq4_xs_q8_K,
|
||||
.vec_dot_type = GGML_TYPE_Q8_K,
|
||||
.nrows = 1,
|
||||
|
|
@ -886,7 +886,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
.is_quantized = false,
|
||||
.to_float = (ggml_to_float_t) ggml_bf16_to_fp32_row,
|
||||
.from_float = (ggml_from_float_t) ggml_fp32_to_bf16_row,
|
||||
.from_float_reference = (ggml_from_float_t) ggml_fp32_to_bf16_row,
|
||||
.from_float_ref = (ggml_from_float_t) ggml_fp32_to_bf16_row,
|
||||
.vec_dot = (ggml_vec_dot_t) ggml_vec_dot_bf16,
|
||||
.vec_dot_type = GGML_TYPE_BF16,
|
||||
.nrows = 1,
|
||||
|
|
@ -894,48 +894,48 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
|
|||
[GGML_TYPE_Q4_0_4_4] = {
|
||||
.type_name = "q4_0_4x4",
|
||||
.blck_size = QK4_0,
|
||||
.blck_size_interleave = 4,
|
||||
.type_size = sizeof(block_q4_0),
|
||||
.is_quantized = true,
|
||||
.to_float = NULL,
|
||||
.from_float = NULL,
|
||||
.from_float_reference = NULL,
|
||||
.from_float_ref = NULL,
|
||||
.vec_dot = NULL,
|
||||
.vec_dot_type = GGML_TYPE_Q8_0,
|
||||
.nrows = 1,
|
||||
.ncols = 4,
|
||||
.interleave_blcksize = 4,
|
||||
.gemv = ggml_gemv_q4_0_4x4_q8_0,
|
||||
.gemm = ggml_gemm_q4_0_4x4_q8_0,
|
||||
},
|
||||
[GGML_TYPE_Q4_0_4_8] = {
|
||||
.type_name = "q4_0_4x8",
|
||||
.blck_size = QK4_0,
|
||||
.blck_size_interleave = 8,
|
||||
.type_size = sizeof(block_q4_0),
|
||||
.is_quantized = true,
|
||||
.to_float = NULL,
|
||||
.from_float = NULL,
|
||||
.from_float_reference = NULL,
|
||||
.from_float_ref = NULL,
|
||||
.vec_dot = NULL,
|
||||
.vec_dot_type = GGML_TYPE_Q8_0,
|
||||
.nrows = 1,
|
||||
.ncols = 4,
|
||||
.interleave_blcksize = 8,
|
||||
.gemv = ggml_gemv_q4_0_4x8_q8_0,
|
||||
.gemm = ggml_gemm_q4_0_4x8_q8_0,
|
||||
},
|
||||
[GGML_TYPE_Q4_0_8_8] = {
|
||||
.type_name = "q4_0_8x8",
|
||||
.blck_size = QK4_0,
|
||||
.blck_size_interleave = 8,
|
||||
.type_size = sizeof(block_q4_0),
|
||||
.is_quantized = true,
|
||||
.to_float = NULL,
|
||||
.from_float = NULL,
|
||||
.from_float_reference = NULL,
|
||||
.from_float_ref = NULL,
|
||||
.vec_dot = NULL,
|
||||
.vec_dot_type = GGML_TYPE_Q8_0,
|
||||
.nrows = 1,
|
||||
.ncols = 8,
|
||||
.interleave_blcksize = 8,
|
||||
.gemv = ggml_gemv_q4_0_8x8_q8_0,
|
||||
.gemm = ggml_gemm_q4_0_8x8_q8_0,
|
||||
}
|
||||
|
|
@ -3115,7 +3115,7 @@ size_t ggml_nbytes_pad(const struct ggml_tensor * tensor) {
|
|||
return GGML_PAD(ggml_nbytes(tensor), GGML_MEM_ALIGN);
|
||||
}
|
||||
|
||||
GGML_CALL int ggml_blck_size(enum ggml_type type) {
|
||||
GGML_CALL int64_t ggml_blck_size(enum ggml_type type) {
|
||||
return type_traits[type].blck_size;
|
||||
}
|
||||
|
||||
|
|
@ -12192,15 +12192,14 @@ static void ggml_compute_forward_mul_mat(
|
|||
|
||||
const enum ggml_type type = src0->type;
|
||||
|
||||
enum ggml_type const vec_dot_type = type_traits[type].vec_dot_type;
|
||||
ggml_from_float_t const from_float_to_vec_dot = type_traits[vec_dot_type].from_float;
|
||||
int64_t const vec_dot_num_rows = type_traits[type].nrows;
|
||||
int64_t const matmul_num_cols = type_traits[type].ncols;
|
||||
int64_t const interleave_blcksize = type_traits[type].interleave_blcksize;
|
||||
ggml_from_float_to_mat_t const from_float_to_mat
|
||||
= type_traits[vec_dot_type].from_float_to_mat;
|
||||
ggml_gemv_t const gemv = type_traits[type].gemv;
|
||||
ggml_gemm_t const gemm = type_traits[type].gemm;
|
||||
enum ggml_type const vec_dot_type = type_traits[type].vec_dot_type;
|
||||
ggml_from_float_t const from_float = type_traits[vec_dot_type].from_float;
|
||||
ggml_from_float_to_mat_t const from_float_to_mat = type_traits[vec_dot_type].from_float_to_mat;
|
||||
int64_t const vec_dot_num_rows = type_traits[type].nrows;
|
||||
int64_t const matmul_num_cols = type_traits[type].ncols;
|
||||
int64_t const blck_size_interleave = type_traits[type].blck_size_interleave;
|
||||
ggml_gemv_t const gemv = type_traits[type].gemv;
|
||||
ggml_gemm_t const gemm = type_traits[type].gemm;
|
||||
|
||||
GGML_ASSERT(ne0 == ne01);
|
||||
GGML_ASSERT(ne1 == ne11);
|
||||
|
|
@ -12264,14 +12263,14 @@ UseGgmlGemm1:;
|
|||
for (int64_t i11 = ith * 4; i11 < ne11 - ne11 % 4; i11 += nth * 4) {
|
||||
from_float_to_mat((float *)((char *) src1->data + i13*nb13 + i12*nb12 + i11*nb11),
|
||||
(void *) (wdata + i13*nbw3 + i12*nbw2 + i11*nbw1),
|
||||
4, ne10, interleave_blcksize);
|
||||
4, ne10, blck_size_interleave);
|
||||
}
|
||||
i11_processed = ne11 - ne11 % 4;
|
||||
}
|
||||
for (int64_t i11 = i11_processed + ith; i11 < ne11; i11 += nth) {
|
||||
from_float_to_vec_dot((float *)((char *) src1->data + i13*nb13 + i12*nb12 + i11*nb11),
|
||||
(void *) (wdata + i13*nbw3 + i12*nbw2 + i11*nbw1),
|
||||
ne10);
|
||||
from_float((float *)((char *) src1->data + i13*nb13 + i12*nb12 + i11*nb11),
|
||||
(void *) (wdata + i13*nbw3 + i12*nbw2 + i11*nbw1),
|
||||
ne10);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
@ -12355,7 +12354,7 @@ UseGgmlGemm2:;
|
|||
int64_t src0_start = (ith * ne01) / nth;
|
||||
int64_t src0_end = ((ith + 1) * ne01) / nth;
|
||||
src0_start = (src0_start % matmul_num_cols) ? src0_start + matmul_num_cols - (src0_start % matmul_num_cols): src0_start;
|
||||
src0_end = (src0_end % matmul_num_cols) ? src0_end + matmul_num_cols - (src0_end % matmul_num_cols): src0_end;
|
||||
src0_end = (src0_end % matmul_num_cols) ? src0_end + matmul_num_cols - (src0_end % matmul_num_cols): src0_end;
|
||||
if (src0_start >= src0_end) return;
|
||||
|
||||
// If there are more than three rows in src1, use gemm; otherwise, use gemv.
|
||||
|
|
@ -12413,11 +12412,11 @@ static void ggml_compute_forward_mul_mat_id(
|
|||
|
||||
const bool src1_cont = ggml_is_contiguous(src1);
|
||||
|
||||
ggml_vec_dot_t const vec_dot = type_traits[type].vec_dot;
|
||||
enum ggml_type const vec_dot_type = type_traits[type].vec_dot_type;
|
||||
ggml_from_float_t const from_float_to_vec_dot = type_traits[vec_dot_type].from_float;
|
||||
int64_t const matmul_num_cols = type_traits[type].ncols;
|
||||
ggml_gemv_t const gemv = type_traits[type].gemv;
|
||||
ggml_vec_dot_t const vec_dot = type_traits[type].vec_dot;
|
||||
enum ggml_type const vec_dot_type = type_traits[type].vec_dot_type;
|
||||
ggml_from_float_t const from_float = type_traits[vec_dot_type].from_float;
|
||||
int64_t const matmul_num_cols = type_traits[type].ncols;
|
||||
ggml_gemv_t const gemv = type_traits[type].gemv;
|
||||
|
||||
// we don't support permuted src0 or src1
|
||||
GGML_ASSERT(nb00 == ggml_type_size(type));
|
||||
|
|
@ -12458,9 +12457,9 @@ static void ggml_compute_forward_mul_mat_id(
|
|||
for (int64_t i13 = 0; i13 < ne13; ++i13) {
|
||||
for (int64_t i12 = 0; i12 < ne12; ++i12) {
|
||||
for (int64_t i11 = ith; i11 < ne11; i11 += nth) {
|
||||
from_float_to_vec_dot((float *)((char *) src1->data + i13*nb13 + i12*nb12 + i11*nb11),
|
||||
(void *) (wdata + i13*nbw3 + i12*nbw2 + i11*nbw1),
|
||||
ne10);
|
||||
from_float((float *)((char *) src1->data + i13*nb13 + i12*nb12 + i11*nb11),
|
||||
(void *) (wdata + i13*nbw3 + i12*nbw2 + i11*nbw1),
|
||||
ne10);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
@ -21063,8 +21062,8 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
|||
(int64_t) info->ne[3];
|
||||
|
||||
if (ne % ggml_blck_size(info->type) != 0) {
|
||||
fprintf(stderr, "%s: tensor '%s' of type %d (%s) number of elements (%" PRId64 ") is not a multiple of block size (%d)\n",
|
||||
__func__, info->name.data, (int)info->type, ggml_type_name(info->type), ne, ggml_blck_size(info->type));
|
||||
fprintf(stderr, "%s: tensor '%s' of type %d (%s) number of elements (%" PRId64 ") is not a multiple of block size (%" PRId64 ")\n",
|
||||
__func__, info->name.data, (int) info->type, ggml_type_name(info->type), ne, ggml_blck_size(info->type));
|
||||
fclose(file);
|
||||
gguf_free(ctx);
|
||||
return NULL;
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue