Skip to content

Commit b268edf

Browse files
committed
llama : bump max layers from 256 to 512
1 parent 5e116e8 commit b268edf

File tree

2 files changed

+3
-3
lines changed

2 files changed

+3
-3
lines changed

include/llama.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -40,7 +40,7 @@
4040
#define LLAMA_FILE_MAGIC_GGSQ 0x67677371u // 'ggsq'
4141

4242
#define LLAMA_SESSION_MAGIC LLAMA_FILE_MAGIC_GGSN
43-
#define LLAMA_SESSION_VERSION 6
43+
#define LLAMA_SESSION_VERSION 7
4444

4545
#define LLAMA_STATE_SEQ_MAGIC LLAMA_FILE_MAGIC_GGSQ
4646
#define LLAMA_STATE_SEQ_VERSION 1

src/llama.cpp

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -112,7 +112,7 @@
112112

113113
// bump if necessary
114114
#define LLAMA_MAX_NODES 8192
115-
#define LLAMA_MAX_LAYERS 256
115+
#define LLAMA_MAX_LAYERS 512
116116
#define LLAMA_MAX_EXPERTS 160 // DeepSeekV2
117117

118118
//
@@ -19877,7 +19877,7 @@ size_t llama_state_get_size(const struct llama_context * ctx) {
1987719877
);
1987819878

1987919879
// on session change it is very likely that the state size has changed - so we need to update this function
19880-
static_assert(LLAMA_SESSION_VERSION == 6, "So you just bumped the session version - good. But did you remember to update llama_state_get_size?");
19880+
static_assert(LLAMA_SESSION_VERSION == 7, "So you just bumped the session version - good. But did you remember to update llama_state_get_size?");
1988119881

1988219882
return s_total;
1988319883
}

0 commit comments

Comments
 (0)