vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : move llama_expert_gating_func_type to llama-hparams.h

Browse source
This commit is contained in:
Stanisław Szymczyk 2025-01-04 17:28:17 +01:00
parent a48c3df3df
commit 4a58b99777
2 changed files with 6 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

6
include/llama.h
View file

@ -116,12 +116,6 @@ extern "C" {
LLAMA_ROPE_TYPE_VISION = GGML_ROPE_TYPE_VISION,
};
enum llama_expert_gating_func_type {
LLAMA_EXPERT_GATING_FUNC_TYPE_NONE = 0,
LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX = 1,
LLAMA_EXPERT_GATING_FUNC_TYPE_SIGMOID = 2,
};
enum llama_token_type { //TODO: remove, required until per token attributes are available from GGUF file
LLAMA_TOKEN_TYPE_UNDEFINED = 0,
LLAMA_TOKEN_TYPE_NORMAL = 1,

6
src/llama-hparams.h
View file

@ -8,6 +8,12 @@
#define LLAMA_MAX_LAYERS 512
#define LLAMA_MAX_EXPERTS 256 // DeepSeekV3
enum llama_expert_gating_func_type {
LLAMA_EXPERT_GATING_FUNC_TYPE_NONE = 0,
LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX = 1,
LLAMA_EXPERT_GATING_FUNC_TYPE_SIGMOID = 2,
};
struct llama_hparams_posnet {
uint32_t n_embd;
uint32_t n_layer;