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

gpt oss harmony template

Browse source
This commit is contained in:
Concedo 2025-08-06 11:39:40 +08:00
parent 6eea7b88d2
commit 34487d3c02
5 changed files with 37 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

10
gpttype_adapter.cpp
View file

@ -2293,15 +2293,21 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
// 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 && inputs.moecpu>0)
if(ggml_backend_dev_count()>1 && inputs.moecpu>0)
{
std::string toadd = "";
for (int i = 0; i < inputs.moecpu; ++i) {
std::string tmp = string_format("blk\\.%d\\.ffn_(up|down|gate)_exps=CPU", i);
if(i>0)
{
tmp = "," + tmp;
}
tensoroverrides += tmp;
toadd += tmp;
}
if (tensoroverrides == "") {
tensoroverrides = toadd;
} else {
tensoroverrides += "," + toadd;
}
printf("Overriding %d MoE layers to CPU...\n",inputs.moecpu);
}