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

llama : bump max layers from 512 to 1024

Browse source
This commit is contained in:
Nico Bosshard 2024-10-16 14:20:20 +02:00
parent 10433e8b45
commit d87aa806b5
2 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

2
include/llama.h
View file

@ -41,7 +41,7 @@
#define LLAMA_FILE_MAGIC_GGSQ 0x67677371u // 'ggsq' #define LLAMA_FILE_MAGIC_GGSQ 0x67677371u // 'ggsq'
#define LLAMA_SESSION_MAGIC LLAMA_FILE_MAGIC_GGSN #define LLAMA_SESSION_MAGIC LLAMA_FILE_MAGIC_GGSN
#define LLAMA_SESSION_VERSION 9 #define LLAMA_SESSION_VERSION 10
#define LLAMA_STATE_SEQ_MAGIC LLAMA_FILE_MAGIC_GGSQ #define LLAMA_STATE_SEQ_MAGIC LLAMA_FILE_MAGIC_GGSQ
#define LLAMA_STATE_SEQ_VERSION 2 #define LLAMA_STATE_SEQ_VERSION 2

2
src/llama.cpp
View file

@ -86,7 +86,7 @@
#endif #endif
// bump if necessary // bump if necessary
#define LLAMA_MAX_LAYERS 512 #define LLAMA_MAX_LAYERS 1024
#define LLAMA_MAX_EXPERTS 160 // DeepSeekV2 #define LLAMA_MAX_EXPERTS 160 // DeepSeekV2
// //