happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

ggml : ggml_set_rows support quantized dst 

ggml-ci
This commit is contained in:
Georgi Gerganov 2025-06-22 11:10:42 +03:00
parent df71c803b4
commit 630c84a2bd
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
2 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/ggml-cpu/ops.cpp
View File

@ -4549,6 +4549,8 @@ static void ggml_compute_forward_set_rows_f32(
const int ir0 = dr*ith; const int ir0 = dr*ith;
const int ir1 = MIN(ir0 + dr, nr); const int ir1 = MIN(ir0 + dr, nr);
ggml_from_float_t const from_float = ggml_get_type_traits_cpu(dst->type)->from_float;
for (int64_t i03 = 0; i03 < ne03; ++i03) { for (int64_t i03 = 0; i03 < ne03; ++i03) {
for (int64_t i02 = 0; i02 < ne02; ++i02) { for (int64_t i02 = 0; i02 < ne02; ++i02) {
for (int64_t i = ir0; i < ir1; ++i) { for (int64_t i = ir0; i < ir1; ++i) {
@ -4560,9 +4562,9 @@ static void ggml_compute_forward_set_rows_f32(
GGML_ASSERT(i01 >= 0 && i01 < ne1); GGML_ASSERT(i01 >= 0 && i01 < ne1);
ggml_cpu_fp32_to_fp16( from_float(
(const float *) ((char *) src0->data + i*nb01 + i02*nb02 + i03*nb03), (const float *) ((char *) src0->data + i*nb01 + i02*nb02 + i03*nb03),
(ggml_fp16_t *) ((char *) dst->data + i01*nb1 + i02*nb2 + i03*nb3), nc); ((char *) dst->data + i01*nb1 + i02*nb2 + i03*nb3), nc);
} }
} }
} }

1
ggml/src/ggml.c
View File

@ -3417,7 +3417,6 @@ struct ggml_tensor * ggml_set_rows(
GGML_ASSERT(b->ne[2] % c->ne[1] == 0); GGML_ASSERT(b->ne[2] % c->ne[1] == 0);
GGML_ASSERT(b->ne[3] % c->ne[2] == 0); GGML_ASSERT(b->ne[3] % c->ne[2] == 0);
GGML_ASSERT(c->ne[3] == 1); GGML_ASSERT(c->ne[3] == 1);
GGML_ASSERT(a->type == GGML_TYPE_F16); // TODO: relax
GGML_ASSERT(b->type == GGML_TYPE_F32); GGML_ASSERT(b->type == GGML_TYPE_F32);
GGML_ASSERT(c->type == GGML_TYPE_I64); GGML_ASSERT(c->type == GGML_TYPE_I64);