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

fix for gemma3n

Browse source
This commit is contained in:
Concedo 2025-07-10 17:39:08 +08:00
parent b8c1fc7c9e
commit c45b8dc56f
4 changed files with 58 additions and 35 deletions
Download patch file Download diff file Expand all files Collapse all files

5
gpttype_adapter.cpp
View file

@ -2287,6 +2287,11 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
}
//handle override tensor
std::string tensoroverrides = inputs.override_tensors;
if(file_format_meta.model_architecture==GGUFArch::ARCH_GEMMA3N)
{
std::string forced = "per_layer_token_embd.weight=CPU"; //this tensor on gpu is problematic on unsloth q4_0
tensoroverrides = (tensoroverrides=="" ? forced: (forced+","+tensoroverrides));
}
if(tensoroverrides!="" && ggml_backend_dev_count()>1)
{
printf("Handling Override Tensors for backends: ");