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

rename n_ctx to kv_size

Browse source
This commit is contained in:
Pierrick HYMBERT 2024-02-18 20:59:26 +01:00 • committed by Georgi Gerganov
parent ef96e8b1f7
commit 606873401c
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
48 changed files with 403 additions and 393 deletions
Download patch file Download diff file Expand all files Collapse all files

22
examples/benchmark/benchmark-matmult.cpp
View file

@ -128,20 +128,20 @@ int main(int argc, char ** argv) {
// TODO: perform the bench for all types or for a user specified type
const ggml_type qtype = GGML_TYPE_Q4_1;
size_t ctx_size = 0;
ctx_size += ggml_row_size(GGML_TYPE_F32, sizex*sizey);
ctx_size += ggml_row_size(GGML_TYPE_F32, sizex*sizey);
ctx_size += ggml_row_size(GGML_TYPE_F32, sizex*sizez);
ctx_size += ggml_row_size(qtype, sizex*sizey);
ctx_size += ggml_row_size(qtype, sizex*sizey);
ctx_size += ggml_row_size(GGML_TYPE_F32, sizex*sizey); // BLAS
ctx_size += ggml_row_size(GGML_TYPE_F32, sizex*sizey); // BLAS
ctx_size += 1024*1024*16;
size_t kv_size = 0;
kv_size += ggml_row_size(GGML_TYPE_F32, sizex * sizey);
kv_size += ggml_row_size(GGML_TYPE_F32, sizex * sizey);
kv_size += ggml_row_size(GGML_TYPE_F32, sizex * sizez);
kv_size += ggml_row_size(qtype, sizex * sizey);
kv_size += ggml_row_size(qtype, sizex * sizey);
kv_size += ggml_row_size(GGML_TYPE_F32, sizex * sizey); // BLAS
kv_size += ggml_row_size(GGML_TYPE_F32, sizex * sizey); // BLAS
kv_size += 1024 * 1024 * 16;
printf("Allocating Memory of size %zi bytes, %zi MB\n",ctx_size, (ctx_size/1024/1024));
printf("Allocating Memory of size %zi bytes, %zi MB\n", kv_size, (kv_size / 1024 / 1024));
struct ggml_init_params params = {
/*.mem_size =*/ ctx_size,
/*.mem_size =*/ kv_size,
/*.mem_buffer =*/ NULL,
/* no_alloc =*/ 0
};