Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 01:05:09 +02:00
Code Issues Packages Projects Releases Wiki Activity

updated kobold lite, patch oom errors

Browse Source
This commit is contained in:
Concedo
2023-05-09 19:16:45 +08:00
parent 6d87f67572
commit e47f7ade05
4 changed files with 13 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
klite.embd
+9 -8
View File
File diff suppressed because one or more lines are too long
otherarch/gpt2_v2.cpp
+2 -2
View File
@@ -350,9 +350,9 @@ 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 + 16u*1024*1024) > buf_size) {
if (mem_per_token > 0 && (mem_per_token*N*2 + 48u*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);
printf("\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new);
// reallocate
if (buf_size_new > buf_size)
otherarch/gptj_v2.cpp
+1 -1
View File
@@ -362,7 +362,7 @@ 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 + 16u*1024*1024) > buf_size) {
if (mem_per_token > 0 && (mem_per_token*N*2 + 48u*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);
otherarch/neox.cpp
+1 -1
View File
@@ -361,7 +361,7 @@ 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 + 16u*1024*1024) > buf_size) {
if (mem_per_token > 0 && (mem_per_token*N*2 + 48u*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);