Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-08-25 22:21:14 +02:00
Code Issues Packages Projects Releases Wiki Activity

gpu layer offloading disabled for phi models in clblast

Browse Source
This commit is contained in:
Concedo
2024-01-25 17:40:05 +08:00
parent 0a70cc1ba7
commit d9a7bd577a
4 changed files with 38 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files
model_adapter.cpp
+10 -12
View File
@@ -255,7 +255,7 @@ void print_tok_vec(std::vector<float> &embd)
else if(magic == 0x46554747)
{
fin.close();
fileformat = FileFormat::GGUF_LLAMA;
fileformat = FileFormat::GGUF_GENERIC;
struct gguf_init_params ggufparams;
ggufparams.no_alloc = true;
@@ -267,19 +267,8 @@ void print_tok_vec(std::vector<float> &embd)
std::string modelarch = "";
if (keyidx != -1) { modelarch = gguf_get_val_str(ctx, keyidx); }
if(modelarch=="llama")
{
fileformat = FileFormat::GGUF_LLAMA;
}
else if(modelarch=="falcon")
{
fileformat = FileFormat::GGUF_FALCON; //uses the same loader
}
printf("\nThe reported GGUF Arch is: %s\n",(modelarch==""?"unknown":modelarch.c_str()));
if(modelarch!="" && fileformatmeta!=nullptr)
{
std::string fkey = modelarch+".context_length";
@@ -289,6 +278,15 @@ void print_tok_vec(std::vector<float> &embd)
}
int filever = gguf_get_version(ctx);
fileformatmeta->fileversion = filever;
fileformatmeta->model_architecture = GGUFArch::DEFAULT;
if(modelarch=="phi2")
{
fileformatmeta->model_architecture = GGUFArch::PHI;
}
else if(modelarch=="falcon")
{
fileformatmeta->model_architecture = GGUFArch::FALCON;
}
}
gguf_free(ctx);
}