vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 01:24:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

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

11
model_adapter.h
View file

@ -21,7 +21,8 @@ enum FileFormat
GGJT=3, // 3=(llama ggjt)
GGJT_2=4, //newer llama format unshuffled
GGJT_3=5, //using 16bit scalar
GGUF_LLAMA=6, //GGUF (llama newest ver)
GGUF_GENERIC=6, //GGUF (llama newest ver)
GPTJ_1=100, //the very first super old GPTJ format
GPTJ_2=101, //pygmalion, uses old ggml lib
@ -47,14 +48,20 @@ enum FileFormat
MPT_1=500, //first supported mpt version
GGUF_FALCON=600, //GGUF (falcon)
};
enum GGUFArch
{
DEFAULT = 0, //used for llama and other generic gguf
FALCON = 1,
PHI = 2,
};
struct FileFormatExtraMeta
{
int n_ctx_train = 2048;
int fileversion = 0;
GGUFArch model_architecture = GGUFArch::DEFAULT;
};
enum ModelLoadResult