vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

imatrix : offload to GPU support (#4957)

Browse source 
* backend : add eval callback

ggml-ci

* backend : group nodes in a single compute when user don't need them

* backend : clean-up the implementation

ggml-ci

* simple : do not perform tensor data copy if not needed

* simple : fix

* imatrix : offload to GPU support

* imatrix : fix ggml_mul_mat_id hanlding

ggml-ci

* ci : add imatrix test

ggml-ci

* ci : rearrange output

ggml-ci
This commit is contained in:
Georgi Gerganov 2024-01-17 18:46:30 +02:00 • committed by GitHub
parent 44a1a4a41a
commit ba69bbc84c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 129 additions and 54 deletions
Download patch file Download diff file Expand all files Collapse all files

14
ggml.c
View file

@ -394,12 +394,6 @@ static const size_t CACHE_LINE_SIZE_F32 = CACHE_LINE_SIZE/sizeof(float);
static void ggml_vec_dot_f32(const int n, float * restrict s, const float * restrict x, const float * restrict y);
static void ggml_vec_dot_f16(const int n, float * restrict s, ggml_fp16_t * restrict x, ggml_fp16_t * restrict y);
ggml_collect_imatrix_t g_imatrix_collect = NULL;
void ggml_set_imatrix_collection(ggml_collect_imatrix_t imatrix_collect) {
g_imatrix_collect = imatrix_collect;
}
static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
[GGML_TYPE_I8] = {
.type_name = "i8",
@ -9790,10 +9784,6 @@ static void ggml_compute_forward_mul_mat(
const int ith = params->ith;
const int nth = params->nth;
if (ith == 1 && g_imatrix_collect) {
g_imatrix_collect(src0, src1);
}
const enum ggml_type type = src0->type;
const bool src1_cont = ggml_is_contiguous(src1);
@ -10097,10 +10087,6 @@ static void ggml_compute_forward_mul_mat_id(
const struct ggml_tensor * src0_cur = dst->src[cur_a + 2];
if (ith == 1 && g_imatrix_collect) {
g_imatrix_collect(src0_cur, src1);
}
const void * wdata = (src1->type == vec_dot_type) ? src1->data : params->wdata;
const size_t row_size = ggml_row_size(vec_dot_type, ne10);