vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-10 09:04:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

unclamp glm4 in debug

Browse source
This commit is contained in:
Concedo 2025-04-30 14:49:52 +08:00
parent 621cc8f33f
commit 9fdec02914
1 changed files with 8 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

6
gpttype_adapter.cpp
View file

@ -1908,10 +1908,16 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
clamped_max_context_length = 16384;
}
if (isGguf && file_format_meta.model_architecture == GGUFArch::ARCH_GLM4 && kcpp_data->n_batch > 16) {
if(debugmode==1)
{
printf("GLM-4 is broken on larger batch sizes. Clamp ignored in debug.\n");
} else {
printf("GLM-4 is broken on larger batch sizes. Clamping batch size to 16.\n");
kcpp_data->n_batch = kcpp_data->n_ubatch = 16;
}
}
kcpp_data->n_ctx = clamped_max_context_length;
max_context_limit_at_load = clamped_max_context_length;
add_bos_token = !inputs.no_bos_token;