feat: sync llama.cpp

scripts/llama.cpp.patch

@@ -1,27 +1,27 @@
-diff --git a/src/llama.cpp/common/common.h b/src/llama.cpp/common/common.h
-index 1d2bd932..b5007c66 100644
---- a/src/llama.cpp/common/common.h
-+++ b/src/llama.cpp/common/common.h
-@@ -183,6 +183,7 @@ struct common_params_vocoder {
- };
-
- struct common_params {
-+    bool vocab_only               = false;
-     int32_t n_predict             =    -1; // new tokens to predict
-     int32_t n_ctx                 =  4096; // context size
-     int32_t n_batch               =  2048; // logical batch size for prompt processing (must be >=32 to use BLAS)
 diff --git a/src/llama.cpp/common/common.cpp b/src/llama.cpp/common/common.cpp
-index 20be9291..1bedc55d 100644
+index 451826d5..a85ac028 100644
 --- a/src/llama.cpp/common/common.cpp
 +++ b/src/llama.cpp/common/common.cpp
-@@ -1017,6 +1017,7 @@ struct llama_model_params common_model_params_to_llama(common_params & params) {
+@@ -1043,6 +1043,7 @@ struct llama_model_params common_model_params_to_llama(common_params & params) {
      if (params.n_gpu_layers != -1) {
          mparams.n_gpu_layers = params.n_gpu_layers;
      }
 +    mparams.vocab_only      = params.vocab_only;
-     mparams.rpc_servers     = params.rpc_servers.c_str();
      mparams.main_gpu        = params.main_gpu;
      mparams.split_mode      = params.split_mode;
+     mparams.tensor_split    = params.tensor_split;
+diff --git a/src/llama.cpp/common/common.h b/src/llama.cpp  /common/common.h
+index 3bcc637c..19ae7dad 100644
+--- a/src/llama.cpp/common/common.h
++++ b/src/llama.cpp/common/common.h
+@@ -189,6 +189,7 @@ struct common_params_vocoder {
+ };
+
+ struct common_params {
++    bool vocab_only               = false;
+     int32_t n_predict             =    -1; // new tokens to predict
+     int32_t n_ctx                 =  4096; // context size
+     int32_t n_batch               =  2048; // logical batch size for prompt processing (must be >=32 to use BLAS)
 diff --git a/src/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt b/src/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt
 index 6b3641c4..6d6cb27f 100644
 --- a/src/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt

src/EmbeddingWorker.cpp

@@ -9,10 +9,11 @@ void EmbeddingWorker::Execute() {
   llama_kv_cache_clear(_sess->context());
   auto tokens = ::common_tokenize(_sess->context(), _text, true);
   // add SEP if not present
-  if (tokens.empty() || tokens.back() != llama_token_sep(_sess->model())) {
-    tokens.push_back(llama_token_sep(_sess->model()));
+  auto vocab = llama_model_get_vocab(_sess->model());
+  if (tokens.empty() || tokens.back() != llama_vocab_sep(vocab)) {
+    tokens.push_back(llama_vocab_sep(vocab));
   }
-  const int n_embd = llama_n_embd(_sess->model());
+  const int n_embd = llama_model_n_embd(_sess->model());
   do {
     auto ctx = _sess->context();
     int ret =

src/LlamaCompletionWorker.cpp

@@ -59,7 +59,9 @@ void LlamaCompletionWorker::Execute() {
   size_t n_cur = 0;
   size_t n_input = 0;
   const auto model = _sess->model();
-  const bool add_bos = llama_add_bos_token(model);
+  auto vocab = llama_model_get_vocab(model);
+
+  const bool add_bos = llama_vocab_get_add_bos(vocab);
   auto ctx = _sess->context();
 
   auto sparams = llama_sampler_chain_default_params();
@@ -130,7 +132,7 @@ void LlamaCompletionWorker::Execute() {
       });
     }
     // is it an end of generation?
-    if (llama_token_is_eog(model, new_token_id)) {
+    if (llama_vocab_is_eog(vocab, new_token_id)) {
       break;
     }
     // check for stop words

src/LlamaContext.cpp

@@ -157,16 +157,10 @@ Napi::Value LlamaContext::GetSystemInfo(const Napi::CallbackInfo &info) {
 }
 
 bool validateModelChatTemplate(const struct llama_model * model) {
-    std::vector<char> model_template(2048, 0); // longest known template is about 1200 bytes
-    std::string template_key = "tokenizer.chat_template";
-    int32_t res = llama_model_meta_val_str(model, template_key.c_str(), model_template.data(), model_template.size());
-    if (res >= 0) {
-        llama_chat_message chat[] = {{"user", "test"}};
-        std::string tmpl = std::string(model_template.data(), model_template.size());
-        int32_t chat_res = llama_chat_apply_template(model, tmpl.c_str(), chat, 1, true, nullptr, 0);
-        return chat_res > 0;
-    }
-    return res > 0;
+  llama_chat_message chat[] = {{"user", "test"}};
+  const char * tmpl = llama_model_chat_template(model);
+  int32_t chat_res = llama_chat_apply_template(tmpl, chat, 1, true, nullptr, 0);
+  return chat_res > 0;
 }
 
 // getModelInfo(): object