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

llama: apply the mllama support patch

Browse source 
Signed-off-by: YiYing He <yiying@secondstate.io>
This commit is contained in:
YiYing He 2025-01-15 17:07:09 +08:00
parent cde3833239
commit 45a89e0cec
16 changed files with 440 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

12
src/llama-model.h
View file

@ -9,6 +9,7 @@
#include <string>
#include <unordered_map>
#include <vector>
#include <stdexcept>
struct llama_model_loader;
@ -62,6 +63,7 @@ enum llm_type {
LLM_TYPE_40B,
LLM_TYPE_65B,
LLM_TYPE_70B,
LLM_TYPE_90B,
LLM_TYPE_236B,
LLM_TYPE_314B,
LLM_TYPE_671B,
@ -281,6 +283,16 @@ struct llama_layer {
struct ggml_tensor * ffn_up_scale = nullptr;
struct ggml_tensor * ffn_down_scale = nullptr;
// cross attention
struct ggml_tensor * cross_attn_k_norm = nullptr;
struct ggml_tensor * cross_attn_k_proj = nullptr;
struct ggml_tensor * cross_attn_o_proj = nullptr;
struct ggml_tensor * cross_attn_q_norm = nullptr;
struct ggml_tensor * cross_attn_q_proj = nullptr;
struct ggml_tensor * cross_attn_v_proj = nullptr;
struct ggml_tensor * cross_attn_attn_gate = nullptr;
struct ggml_tensor * cross_attn_mlp_gate = nullptr;
struct llama_layer_posnet posnet;
struct llama_layer_convnext convnext;