From b268edf87c3bb996feb88e2d8b2c85b38bff6f36 Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Wed, 17 Jul 2024 10:01:54 +0300 Subject: [PATCH] llama : bump max layers from 256 to 512 --- include/llama.h | 2 +- src/llama.cpp | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/include/llama.h b/include/llama.h index c0fb53060..b280df325 100644 --- a/include/llama.h +++ b/include/llama.h @@ -40,7 +40,7 @@ #define LLAMA_FILE_MAGIC_GGSQ 0x67677371u // 'ggsq' #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_VERSION 1 diff --git a/src/llama.cpp b/src/llama.cpp index 4a9903cc3..0a34f46e6 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -112,7 +112,7 @@ // bump if necessary #define LLAMA_MAX_NODES 8192 -#define LLAMA_MAX_LAYERS 256 +#define LLAMA_MAX_LAYERS 512 #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 - 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; }