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

inc allocation

Browse source
This commit is contained in:
Concedo 2023-05-16 23:32:35 +08:00
parent 954b87eb05
commit 94ef3e81cf
3 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

2
otherarch/gpt2_v2.cpp
View file

@ -352,7 +352,7 @@ bool gpt2_eval(
static size_t buf_size = 256u*1024*1024;
static void * buf = malloc(buf_size);
if (mem_per_token > 0 && (mem_per_token*N*2 + 48u*1024*1024) > buf_size) {
if (mem_per_token > 0 && (mem_per_token*N*2 + 64u*1024*1024) > buf_size) {
const size_t buf_size_new = 320u*1024*1024 + 2*(mem_per_token*N); // add 10% to account for ggml object overhead
//printf("\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new);

2
otherarch/gptj_v2.cpp
View file

@ -364,7 +364,7 @@ bool gptj_eval(
static size_t buf_size = 256u*1024*1024;
static void * buf = malloc(buf_size);
if (mem_per_token > 0 && (mem_per_token*N*2 + 48u*1024*1024) > buf_size) {
if (mem_per_token > 0 && (mem_per_token*N*2 + 64u*1024*1024) > buf_size) {
const size_t buf_size_new = 320u*1024*1024 + 2*(mem_per_token*N); // add 10% to account for ggml object overhead
//printf("\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new);

2
otherarch/neox.cpp
View file

@ -362,7 +362,7 @@ bool stablelm_eval(
static size_t buf_size = 256u*1024*1024;
static void * buf = malloc(buf_size);
if (mem_per_token > 0 && (mem_per_token*N*2 + 48u*1024*1024) > buf_size) {
if (mem_per_token > 0 && (mem_per_token*N*2 + 64u*1024*1024) > buf_size) {
const size_t buf_size_new = 360u*1024*1024 + 2*(mem_per_token*N); // add 10% to account for ggml object overhead
//printf("\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new);