llama.cpp updated to b2692

Package.swift

@@ -23,16 +23,16 @@ let package = Package(
             name: "llmfarm_core_cpp",
             sources: ["ggml/ggml.c","exception_helper.cpp","ggml/ggml-quants.c","ggml/ggml-alloc.c","ggml/ggml-backend.c","ggml/ggml-metal.m",
                       "ggml/common.cpp","ggml/sampling.cpp","ggml/train.cpp","ggml/build-info.cpp",
-                      "gpt_helpers.cpp","gpt_spm.cpp","package_helper.m","grammar-parser.cpp","exception_helper_objc.mm","ggml/common_old.cpp",
-                      "llava/llava.cpp","llava/clip.cpp",
-                      "llava/llava-cli.cpp",
-                      "finetune/finetune.cpp","finetune/export-lora.cpp","llama/llama.cpp",
+                      "gpt_helpers.cpp","gpt_spm.cpp","package_helper.m","ggml/grammar-parser.cpp","exception_helper_objc.mm","ggml/common_old.cpp",
+                      "llava/llava.cpp","llava/clip.cpp","llava/llava-cli.cpp","llama/unicode.cpp","llama/unicode-data.cpp","ggml/sgemm.cpp",
+                      "ggml/json-schema-to-grammar.cpp","finetune/finetune.cpp","finetune/export-lora.cpp","llama/llama.cpp",
                       "ggml/ggml_d925ed.c","ggml/ggml_d925ed-alloc.c","ggml/ggml_d925ed-metal.m","rwkv/rwkv.cpp",
                       "ggml/ggml_dadbed9.c","ggml/k_quants_dadbed9.c","ggml/ggml-alloc_dadbed9.c","ggml/ggml-metal_dadbed9.m",
                       "gptneox/gptneox.cpp","gpt2/gpt2.cpp","replit/replit.cpp","starcoder/starcoder.cpp","llama/llama_dadbed9.cpp"
                       ],
             resources: [
                 .copy("tokenizers"),
+                .process("ggml-metal.metal"),
                 .copy("metal")
             ],
             publicHeadersPath: "spm-headers",

Sources/llmfarm_core/LLaMa.swift

@@ -107,7 +107,7 @@ public class LLaMa: LLMBase {
             model_params.use_mmap = false
         }
 
-        llama_backend_init(false)
+        llama_backend_init()
 
         self.model = llama_load_model_from_file(path, model_params)
         if self.model == nil{

Sources/llmfarm_core/LLaMa_MModal.swift

@@ -20,7 +20,7 @@ public class LLaMa_MModal: LLaMa {
         #if os(iOS)
         self.clip_ctx = clip_model_load(contextParams.clip_model, 1, 0 );
         #else
-        self.clip_ctx = clip_model_load(contextParams.clip_model, 1,contextParams.use_metal ? 1: 0 );
+        self.clip_ctx = clip_model_load(contextParams.clip_model, 1,contextParams.use_metal ? 1: 0);
         #endif
         return true
     }

Sources/llmfarm_core_cpp/finetune/export-lora.cpp

@@ -8,8 +8,6 @@
 #include <string>
 #include <thread>
 
-static const size_t tensor_alignment = 32;
-
 struct lora_info {
     std::string filename;
     float scale;