llama : bump max layers from 256 to 512 (#8530)

* llama : bump max layers from 256 to 512

* llama : replace asserts with exceptions

README.md

@@ -3,7 +3,7 @@
 ![llama](https://user-images.githubusercontent.com/1991296/230134379-7181e485-c521-4d23-a0d6-f7b3b61ba524.png)
 
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
-[![Server](https://github.com/ggerganov/llama.cpp/actions/workflows/server.yml/badge.svg?branch=master&event=schedule)](https://github.com/ggerganov/llama.cpp/actions/workflows/server.yml)
+[![Server](https://github.com/ggerganov/llama.cpp/actions/workflows/server.yml/badge.svg)](https://github.com/ggerganov/llama.cpp/actions/workflows/server.yml)
 [![Conan Center](https://shields.io/conan/v/llama-cpp)](https://conan.io/center/llama-cpp)
 
 [Roadmap](https://github.com/users/ggerganov/projects/7) / [Project status](https://github.com/ggerganov/llama.cpp/discussions/3471) / [Manifesto](https://github.com/ggerganov/llama.cpp/discussions/205) / [ggml](https://github.com/ggerganov/ggml)

ggml/src/ggml.c

@@ -19019,7 +19019,7 @@ void ggml_graph_export(const struct ggml_cgraph * cgraph, const char * fname) {
         FILE * fout = ggml_fopen(fname, "wb");
 
         if (!fout) {
-            fprintf(stderr, "%s: failed to open %s\n", __func__, fname);
+            fprintf(stderr, "%s: failed to open %s: %s\n", __func__, fname, strerror(errno));
             return;
         }
 
@@ -19156,7 +19156,7 @@ struct ggml_cgraph * ggml_graph_import(const char * fname, struct ggml_context *
     {
         FILE * fin = ggml_fopen(fname, "rb");
         if (!fin) {
-            fprintf(stderr, "%s: failed to open %s\n", __func__, fname);
+            fprintf(stderr, "%s: failed to open %s: %s\n", __func__, fname, strerror(errno));
             return result;
         }
 
@@ -20830,6 +20830,7 @@ struct gguf_context * gguf_init_empty(void) {
 struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_params params) {
     FILE * file = ggml_fopen(fname, "rb");
     if (!file) {
+        fprintf(stderr, "%s: failed to open '%s': '%s'\n", __func__, fname, strerror(errno));
         return NULL;
     }
 

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

src/llama.cpp

@@ -114,7 +114,7 @@
 
 // bump if necessary
 #define LLAMA_MAX_NODES   8192
-#define LLAMA_MAX_LAYERS  256
+#define LLAMA_MAX_LAYERS  512
 #define LLAMA_MAX_EXPERTS 160  // DeepSeekV2
 
 //
@@ -4007,7 +4007,9 @@ struct llama_model_loader {
                 throw std::runtime_error(format("%s is not a float32, int32 array", key.c_str()));
         }
 
-        GGML_ASSERT(arr_info.length <= N_MAX);
+        if (arr_info.length > N_MAX) {
+            throw std::runtime_error(format("array length %u for key %s exceeds max %u", (uint32_t) arr_info.length, key.c_str(), (uint32_t) N_MAX));
+        }
 
         std::copy((const T*)arr_info.data, (const T *)arr_info.data + arr_info.length, result.begin());
 
@@ -4043,8 +4045,6 @@ struct llama_model_loader {
     // get array of n <= N_MAX elements, or a single element repeated n times
     template<typename T, size_t N_MAX>
     bool get_key_or_arr(const std::string & key, std::array<T, N_MAX> & result, uint32_t n, const bool required = true) {
-        GGML_ASSERT(n <= N_MAX);
-
         const int kid = gguf_find_key(meta, key.c_str());
 
         if (kid < 0) {
@@ -4054,6 +4054,10 @@ struct llama_model_loader {
             return false;
         }
 
+        if (n > N_MAX) {
+            throw std::runtime_error(format("n > N_MAX: %u > %u for key %s", (uint32_t) n, (uint32_t) N_MAX, key.c_str()));
+        }
+
         if (gguf_get_kv_type(meta, kid) == GGUF_TYPE_ARRAY) {
             struct GGUFMeta::ArrayInfo arr_info =
                 GGUFMeta::GKV<GGUFMeta::ArrayInfo>::get_kv(meta, kid);
@@ -19920,7 +19924,7 @@ 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;
 }
@@ -21607,7 +21611,7 @@ static int32_t llama_chat_apply_template_internal(
         if (add_ass) {
             ss << "<|assistant|>";
         }
-    } else if (tmpl == "chaglm4" || tmpl_contains("[gMASK]<sop>")) {
+    } else if (tmpl == "chatglm4" || tmpl_contains("[gMASK]<sop>")) {
         ss << "[gMASK]" << "<sop>";
         for (auto message : chat) {
             std::string role(message->role);