llama : bump max layers from 256 to 512
This commit is contained in:
parent
5e116e8dd5
commit
b268edf87c
2 changed files with 3 additions and 3 deletions
|
@ -40,7 +40,7 @@
|
||||||
#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 6
|
#define LLAMA_SESSION_VERSION 7
|
||||||
|
|
||||||
#define LLAMA_STATE_SEQ_MAGIC LLAMA_FILE_MAGIC_GGSQ
|
#define LLAMA_STATE_SEQ_MAGIC LLAMA_FILE_MAGIC_GGSQ
|
||||||
#define LLAMA_STATE_SEQ_VERSION 1
|
#define LLAMA_STATE_SEQ_VERSION 1
|
||||||
|
|
|
@ -112,7 +112,7 @@
|
||||||
|
|
||||||
// bump if necessary
|
// bump if necessary
|
||||||
#define LLAMA_MAX_NODES 8192
|
#define LLAMA_MAX_NODES 8192
|
||||||
#define LLAMA_MAX_LAYERS 256
|
#define LLAMA_MAX_LAYERS 512
|
||||||
#define LLAMA_MAX_EXPERTS 160 // DeepSeekV2
|
#define LLAMA_MAX_EXPERTS 160 // DeepSeekV2
|
||||||
|
|
||||||
//
|
//
|
||||||
|
@ -19877,7 +19877,7 @@ size_t llama_state_get_size(const struct llama_context * ctx) {
|
||||||
);
|
);
|
||||||
|
|
||||||
// on session change it is very likely that the state size has changed - so we need to update this function
|
// on session change it is very likely that the state size has changed - so we need to update this function
|
||||||
static_assert(LLAMA_SESSION_VERSION == 6, "So you just bumped the session version - good. But did you remember to update llama_state_get_size?");
|
static_assert(LLAMA_SESSION_VERSION == 7, "So you just bumped the session version - good. But did you remember to update llama_state_get_size?");
|
||||||
|
|
||||||
return s_total;
|
return s_total;
|
||||||
}
|
}
|
||||||
|
|
Loading…
Add table
Add a link
Reference in a new issue