Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 09:15:18 +02:00
Code Issues Packages Projects Releases Wiki Activity

still needs to be a bit higher for very small contexts

Browse Source
This commit is contained in:
Concedo
2023-04-23 15:01:38 +08:00
parent 4e1ea2ac61
commit 432cc91649
3 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
otherarch/gpt2_v2.cpp
+2 -2
View File
@@ -374,8 +374,8 @@ 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*1.6 > buf_size) {
const size_t buf_size_new = 2*(mem_per_token*N); // add 10% to account for ggml object overhead
if (mem_per_token > 0 && mem_per_token*N*1.9 > 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);
// reallocate
otherarch/gptj_v2.cpp
+2 -2
View File
@@ -385,8 +385,8 @@ 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*1.6 > buf_size) {
const size_t buf_size_new = 1.8*(mem_per_token*N); // add 10% to account for ggml object overhead
if (mem_per_token > 0 && mem_per_token*N*1.9 > 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);
// reallocate
otherarch/neox.cpp
+2 -2
View File
@@ -367,8 +367,8 @@ 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*1.6 > buf_size) {
const size_t buf_size_new = 1.8*(mem_per_token*N); // add 10% to account for ggml object overhead
if (mem_per_token > 0 && mem_per_token*N*1.9 > 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);
// reallocate