@fugood/llama.node 0.3.13 → 0.3.15

package/lib/binding.ts

@@ -87,7 +87,7 @@ export type LlamaCompletionOptions = {
   stop?: string[]
   grammar?: string
   grammar_lazy?: boolean
-  grammar_triggers?: { word: string; at_start: boolean }[]
+  grammar_triggers?: { type: number; word: string; at_start: boolean }[]
   preserved_tokens?: string[]
 }
 

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@fugood/llama.node",
   "access": "public",
-  "version": "0.3.13",
+  "version": "0.3.15",
   "description": "An another Node binding of llama.cpp",
   "main": "lib/index.js",
   "scripts": {

package/src/LlamaContext.cpp

@@ -272,7 +272,7 @@ LlamaContext::LlamaContext(const Napi::CallbackInfo &info)
   _sess = sess;
   _info = common_params_get_system_info(params);
 
-  _templates = common_chat_templates_from_model(model, params.chat_template);
+  _templates = common_chat_templates_init(model, params.chat_template);
 }
 
 // getSystemInfo(): string
@@ -355,22 +355,22 @@ Napi::Value LlamaContext::GetModelInfo(const Napi::CallbackInfo &info) {
   Napi::Object minja = Napi::Object::New(info.Env());
   minja.Set("default", validateModelChatTemplate(model, true, ""));
   Napi::Object defaultCaps = Napi::Object::New(info.Env());
-  defaultCaps.Set("tools", _templates.template_default->original_caps().supports_tools);
-  defaultCaps.Set("toolCalls", _templates.template_default->original_caps().supports_tool_calls);
-  defaultCaps.Set("toolResponses", _templates.template_default->original_caps().supports_tool_responses);
-  defaultCaps.Set("systemRole", _templates.template_default->original_caps().supports_system_role);
-  defaultCaps.Set("parallelToolCalls", _templates.template_default->original_caps().supports_parallel_tool_calls);
-  defaultCaps.Set("toolCallId", _templates.template_default->original_caps().supports_tool_call_id);
+  defaultCaps.Set("tools", _templates.get()->template_default->original_caps().supports_tools);
+  defaultCaps.Set("toolCalls", _templates.get()->template_default->original_caps().supports_tool_calls);
+  defaultCaps.Set("toolResponses", _templates.get()->template_default->original_caps().supports_tool_responses);
+  defaultCaps.Set("systemRole", _templates.get()->template_default->original_caps().supports_system_role);
+  defaultCaps.Set("parallelToolCalls", _templates.get()->template_default->original_caps().supports_parallel_tool_calls);
+  defaultCaps.Set("toolCallId", _templates.get()->template_default->original_caps().supports_tool_call_id);
   minja.Set("defaultCaps", defaultCaps);
   minja.Set("toolUse", validateModelChatTemplate(model, true, "tool_use"));
-  if (_templates.template_tool_use) {
+  if (_templates.get()->template_tool_use) {
     Napi::Object toolUseCaps = Napi::Object::New(info.Env());
-    toolUseCaps.Set("tools", _templates.template_tool_use->original_caps().supports_tools);
-    toolUseCaps.Set("toolCalls", _templates.template_tool_use->original_caps().supports_tool_calls);
-    toolUseCaps.Set("toolResponses", _templates.template_tool_use->original_caps().supports_tool_responses);
-    toolUseCaps.Set("systemRole", _templates.template_tool_use->original_caps().supports_system_role);
-    toolUseCaps.Set("parallelToolCalls", _templates.template_tool_use->original_caps().supports_parallel_tool_calls);
-    toolUseCaps.Set("toolCallId", _templates.template_tool_use->original_caps().supports_tool_call_id);
+    toolUseCaps.Set("tools", _templates.get()->template_tool_use->original_caps().supports_tools);
+    toolUseCaps.Set("toolCalls", _templates.get()->template_tool_use->original_caps().supports_tool_calls);
+    toolUseCaps.Set("toolResponses", _templates.get()->template_tool_use->original_caps().supports_tool_responses);
+    toolUseCaps.Set("systemRole", _templates.get()->template_tool_use->original_caps().supports_system_role);
+    toolUseCaps.Set("parallelToolCalls", _templates.get()->template_tool_use->original_caps().supports_parallel_tool_calls);
+    toolUseCaps.Set("toolCallId", _templates.get()->template_tool_use->original_caps().supports_tool_call_id);
     minja.Set("toolUseCaps", toolUseCaps);
   }
   chatTemplates.Set("minja", minja);
@@ -385,7 +385,7 @@ Napi::Value LlamaContext::GetModelInfo(const Napi::CallbackInfo &info) {
 
 common_chat_params getFormattedChatWithJinja(
   const std::shared_ptr<LlamaSession> &sess,
-  const common_chat_templates &templates,
+  const common_chat_templates_ptr &templates,
   const std::string &messages,
   const std::string &chat_template,
   const std::string &json_schema,
@@ -393,72 +393,46 @@ common_chat_params getFormattedChatWithJinja(
   const bool &parallel_tool_calls,
   const std::string &tool_choice
 ) {
-  common_chat_inputs inputs;
-  inputs.messages = json::parse(messages);
+  common_chat_templates_inputs inputs;
+  inputs.messages = common_chat_msgs_parse_oaicompat(json::parse(messages));
   auto useTools = !tools.empty();
   if (useTools) {
-      inputs.tools = json::parse(tools);
+    inputs.tools = common_chat_tools_parse_oaicompat(json::parse(tools));
   }
   inputs.parallel_tool_calls = parallel_tool_calls;
   if (!tool_choice.empty()) {
-      inputs.tool_choice = tool_choice;
+    inputs.tool_choice = common_chat_tool_choice_parse_oaicompat(tool_choice);
   }
   if (!json_schema.empty()) {
-      inputs.json_schema = json::parse(json_schema);
+    inputs.json_schema = json::parse(json_schema);
   }
   inputs.extract_reasoning = sess->params().reasoning_format != COMMON_REASONING_FORMAT_NONE;
-  inputs.stream = true;
 
   // If chat_template is provided, create new one and use it (probably slow)
   if (!chat_template.empty()) {
-      auto tmp = common_chat_templates_from_model(sess->model(), chat_template);
-      const common_chat_template* template_ptr = useTools && tmp.template_tool_use ? tmp.template_tool_use.get() : tmp.template_default.get();
-      if (inputs.parallel_tool_calls && !template_ptr->original_caps().supports_parallel_tool_calls) {
-          inputs.parallel_tool_calls = false;
-      }
-      return common_chat_params_init(*template_ptr, inputs);
+      auto tmps = common_chat_templates_init(sess->model(), chat_template);
+      return common_chat_templates_apply(tmps.get(), inputs);
   } else {
-      const common_chat_template* template_ptr = useTools && templates.template_tool_use ? templates.template_tool_use.get() : templates.template_default.get();
-      if (inputs.parallel_tool_calls && !template_ptr->original_caps().supports_parallel_tool_calls) {
-          inputs.parallel_tool_calls = false;
-      }
-      return common_chat_params_init(*template_ptr, inputs);
+      return common_chat_templates_apply(templates.get(), inputs);
   }
 }
 
 std::string getFormattedChat(
   const struct llama_model * model,
-  const common_chat_templates &templates,
+  const common_chat_templates_ptr &templates,
   const std::string &messages,
   const std::string &chat_template
 ) {
-  auto chat_json = json::parse(messages);
-
-  // Handle regular chat without tools
-  std::vector<common_chat_msg> chat_msgs;
-  for (const auto &msg : chat_json) {
-      chat_msgs.push_back({
-          msg["role"].get<std::string>(),
-          msg["content"].get<std::string>()
-      });
-  }
+  common_chat_templates_inputs inputs;
+  inputs.messages = common_chat_msgs_parse_oaicompat(json::parse(messages));
+  inputs.use_jinja = false;
 
   // If chat_template is provided, create new one and use it (probably slow)
   if (!chat_template.empty()) {
-      auto tmp = common_chat_templates_from_model(model, chat_template);
-      return common_chat_apply_template(
-          *tmp.template_default,
-          chat_msgs,
-          true,
-          false
-      );
+    auto tmps = common_chat_templates_init(model, chat_template);
+    return common_chat_templates_apply(tmps.get(), inputs).prompt;
   } else {
-      return common_chat_apply_template(
-          *templates.template_default,
-          chat_msgs,
-          true,
-          false
-      );
+    return common_chat_templates_apply(templates.get(), inputs).prompt;
   }
 }
 
@@ -504,20 +478,21 @@ Napi::Value LlamaContext::GetFormattedChat(const Napi::CallbackInfo &info) {
     auto chatParams = getFormattedChatWithJinja(_sess, _templates, messages, chat_template, json_schema_str, tools_str, parallel_tool_calls, tool_choice);
     
     Napi::Object result = Napi::Object::New(env);
-    result.Set("prompt", chatParams.prompt.get<std::string>());
+    result.Set("prompt", chatParams.prompt);
     // chat_format: int
     result.Set("chat_format", static_cast<int>(chatParams.format));
     // grammar: string
     result.Set("grammar", chatParams.grammar);
     // grammar_lazy: boolean
     result.Set("grammea_lazy", chatParams.grammar_lazy);
-    // grammar_triggers: [{ word: string, at_start: boolean }]
+    // grammar_triggers: [{ value: string, token: number }]
     Napi::Array grammar_triggers = Napi::Array::New(env);
     for (size_t i = 0; i < chatParams.grammar_triggers.size(); i++) {
         const auto & trigger = chatParams.grammar_triggers[i];
         Napi::Object triggerObj = Napi::Object::New(env);
-        triggerObj.Set("word", Napi::String::New(env, trigger.word.c_str()));
-        triggerObj.Set("at_start", Napi::Boolean::New(env, trigger.at_start));
+        triggerObj.Set("type", Napi::Number::New(env, trigger.type));
+        triggerObj.Set("value", Napi::String::New(env, trigger.value));
+        triggerObj.Set("token", Napi::Number::New(env, trigger.token));
         grammar_triggers.Set(i, triggerObj);
     }
     result.Set("grammar_triggers", grammar_triggers);
@@ -594,6 +569,60 @@ Napi::Value LlamaContext::Completion(const Napi::CallbackInfo &info) {
     }
   }
 
+  // Handle preserved_tokens from options
+  if (options.Has("preserved_tokens")) {
+    auto preserved_tokens = options.Get("preserved_tokens").As<Napi::Array>();
+    for (size_t i = 0; i < preserved_tokens.Length(); i++) {
+      auto token = preserved_tokens.Get(i).ToString().Utf8Value();
+      auto ids = common_tokenize(_sess->context(), token, /* add_special= */ false, /* parse_special= */ true);
+      if (ids.size() == 1) {
+        params.sampling.preserved_tokens.insert(ids[0]);
+      }
+    }
+  }
+
+  // Handle grammar_triggers from options
+  if (options.Has("grammar_triggers")) {
+    auto grammar_triggers = options.Get("grammar_triggers").As<Napi::Array>();
+    for (size_t i = 0; i < grammar_triggers.Length(); i++) {
+      auto trigger_obj = grammar_triggers.Get(i).As<Napi::Object>();
+
+      auto type = static_cast<common_grammar_trigger_type>(trigger_obj.Get("type").ToNumber().Int32Value());
+      auto word = trigger_obj.Get("value").ToString().Utf8Value();
+
+      if (type == COMMON_GRAMMAR_TRIGGER_TYPE_WORD) {
+        auto ids = common_tokenize(_sess->context(), word, /* add_special= */ false, /* parse_special= */ true);
+        if (ids.size() == 1) {
+          auto token = ids[0];
+          if (std::find(params.sampling.preserved_tokens.begin(), params.sampling.preserved_tokens.end(), (llama_token) token) == params.sampling.preserved_tokens.end()) {
+            throw std::runtime_error("Grammar trigger word should be marked as preserved token");
+          }
+          common_grammar_trigger trigger;
+          trigger.type = COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN;
+          trigger.value = word;
+          trigger.token = token;
+          params.sampling.grammar_triggers.push_back(std::move(trigger));
+        } else {
+          params.sampling.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, word});
+        }
+      } else {
+        common_grammar_trigger trigger;
+        trigger.type = type;
+        trigger.value = word;
+        if (type == COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN) {
+          auto token = (llama_token) trigger_obj.Get("token").ToNumber().Int32Value();
+          trigger.token = token;
+        }
+        params.sampling.grammar_triggers.push_back(std::move(trigger));
+      }
+    }
+  }
+
+  // Handle grammar_lazy from options
+  if (options.Has("grammar_lazy")) {
+    params.sampling.grammar_lazy = options.Get("grammar_lazy").ToBoolean().Value();
+  }
+
   if (options.Has("messages") && options.Get("messages").IsArray()) {
     auto messages = options.Get("messages").As<Napi::Array>();
     auto chat_template = get_option<std::string>(options, "chat_template", "");
@@ -616,33 +645,26 @@ Napi::Value LlamaContext::Completion(const Napi::CallbackInfo &info) {
         tool_choice
       );  
       
-      params.prompt = chatParams.prompt.get<std::string>();
+      params.prompt = chatParams.prompt;
 
       chat_format = chatParams.format;
 
+      for (const auto & token : chatParams.preserved_tokens) {
+        auto ids = common_tokenize(_sess->context(), token, /* add_special= */ false, /* parse_special= */ true);
+        if (ids.size() == 1) {
+          params.sampling.preserved_tokens.insert(ids[0]);
+        }
+      }
+
       if (!has_grammar_set) {
         // grammar param always wins jinja template & json_schema
         params.sampling.grammar = chatParams.grammar;
         params.sampling.grammar_lazy = chatParams.grammar_lazy;
-        
         for (const auto & trigger : chatParams.grammar_triggers) {
-          auto ids = common_tokenize(_sess->context(), trigger.word, /* add_special= */ false, /* parse_special= */ true);
-          if (ids.size() == 1) {
-            params.sampling.grammar_trigger_tokens.push_back(ids[0]);
-            params.sampling.preserved_tokens.insert(ids[0]);
-            continue;
-          }
-          params.sampling.grammar_trigger_words.push_back(trigger);
+          params.sampling.grammar_triggers.push_back(trigger);
         }
         has_grammar_set = true;
       }
-
-      for (const auto & token : chatParams.preserved_tokens) {
-        auto ids = common_tokenize(_sess->context(), token, /* add_special= */ false, /* parse_special= */ true);
-        if (ids.size() == 1) {
-          params.sampling.preserved_tokens.insert(ids[0]);
-        }
-      }
       
       for (const auto & stop : chatParams.additional_stops) {
         stop_words.push_back(stop);

package/src/LlamaContext.h

@@ -28,7 +28,7 @@ private:
   std::string _info;
   Napi::Object _meta;
   LlamaSessionPtr _sess = nullptr;
-  common_chat_templates _templates;
+  common_chat_templates_ptr _templates;
   std::vector<common_adapter_lora_info> _lora;
   LlamaCompletionWorker *_wip = nullptr;
 };

package/src/common.hpp

@@ -2,8 +2,7 @@
 
 #include "common/common.h"
 #include "common/sampling.h"
-#include "chat.hpp"
-#include "chat-template.hpp"
+#include "chat.h"
 #include "llama.h"
 #include <memory>
 #include <mutex>

package/src/llama.cpp/.github/workflows/build.yml

@@ -173,7 +173,15 @@ jobs:
           name: llama-bin-macos-x64.zip
 
   ubuntu-cpu-cmake:
-    runs-on: ubuntu-22.04
+    strategy:
+      matrix:
+        include:
+          - build: 'x64'
+            os: ubuntu-22.04
+          - build: 'arm64'
+            os: ubuntu-22.04-arm
+
+    runs-on: ${{ matrix.os }}
 
     steps:
       - name: Clone
@@ -239,14 +247,14 @@ jobs:
         run: |
           cp LICENSE ./build/bin/
           cp examples/run/linenoise.cpp/LICENSE ./build/bin/LICENSE.linenoise.cpp
-          zip -r llama-${{ steps.tag.outputs.name }}-bin-ubuntu-x64.zip ./build/bin/*
+          zip -r llama-${{ steps.tag.outputs.name }}-bin-ubuntu-${{ matrix.build }}.zip ./build/bin/*
 
       - name: Upload artifacts
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
         uses: actions/upload-artifact@v4
         with:
-          path: llama-${{ steps.tag.outputs.name }}-bin-ubuntu-x64.zip
-          name: llama-bin-ubuntu-x64.zip
+          path: llama-${{ steps.tag.outputs.name }}-bin-ubuntu-${{ matrix.build }}.zip
+          name: llama-bin-ubuntu-${{ matrix.build }}.zip
 
   ubuntu-latest-cmake-sanitizer:
     runs-on: ubuntu-latest
@@ -459,6 +467,7 @@ jobs:
         run: |
           cmake -B build -S . \
             -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" \
+            -DGGML_HIP_ROCWMMA_FATTN=ON \
             -DGGML_HIP=ON
           cmake --build build --config Release -j $(nproc)
 
@@ -468,6 +477,7 @@ jobs:
           cmake -B build2 -S . \
             -DCMAKE_C_COMPILER=hipcc \
             -DCMAKE_CXX_COMPILER=hipcc \
+            -DGGML_HIP_ROCWMMA_FATTN=ON \
             -DGGML_HIP=ON
           cmake --build build2 --config Release -j $(nproc)
 
@@ -666,6 +676,35 @@ jobs:
             -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
           cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
 
+  macOS-latest-cmake-visionos:
+    runs-on: macos-latest
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v4
+
+      - name: Dependencies
+        id: depends
+        continue-on-error: true
+        run: |
+          brew update
+
+      - name: Build
+        id: cmake_build
+        run: |
+          sysctl -a
+          cmake -B build -G Xcode \
+            -DGGML_METAL_USE_BF16=ON \
+            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_BUILD_EXAMPLES=OFF \
+            -DLLAMA_BUILD_TESTS=OFF \
+            -DLLAMA_BUILD_SERVER=OFF \
+            -DCMAKE_SYSTEM_NAME=visionOS \
+            -DCMAKE_OSX_DEPLOYMENT_TARGET=1.0 \
+            -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
+          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
+
   macOS-latest-swift:
     runs-on: macos-latest
 
@@ -702,12 +741,11 @@ jobs:
             -DLLAMA_BUILD_SERVER=OFF \
             -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64"
           cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
-          sudo cmake --install build --config Release
 
       - name: xcodebuild for swift package
         id: xcodebuild
         run: |
-          xcodebuild -scheme llama-Package -destination "${{ matrix.destination }}"
+          ./build-xcframework.sh
 
   windows-msys2:
     runs-on: windows-latest
@@ -765,7 +803,7 @@ jobs:
     env:
       OPENBLAS_VERSION: 0.3.23
       SDE_VERSION: 9.33.0-2024-01-07
-      VULKAN_VERSION: 1.3.261.1
+      VULKAN_VERSION: 1.4.304.1
 
     strategy:
       matrix:
@@ -1195,6 +1233,11 @@ jobs:
         id: checkout
         uses: actions/checkout@v4
 
+      - name: Clone rocWMMA repository
+        id: clone_rocwmma
+        run: |
+          git clone https://github.com/rocm/rocwmma --branch rocm-6.2.4 --depth 1
+
       - name: Install
         id: depends
         run: |
@@ -1224,8 +1267,10 @@ jobs:
           cmake -G "Unix Makefiles" -B build -S . `
             -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" `
             -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" `
+            -DCMAKE_CXX_FLAGS="-I$($PWD.Path.Replace('\', '/'))/rocwmma/library/include/" `
             -DCMAKE_BUILD_TYPE=Release `
             -DGGML_HIP=ON `
+            -DGGML_HIP_ROCWMMA_FATTN=ON `
             -DGGML_RPC=ON
           cmake --build build -j ${env:NUMBER_OF_PROCESSORS}
 
@@ -1244,6 +1289,11 @@ jobs:
         with:
             fetch-depth: 0
 
+      - name: Clone rocWMMA repository
+        id: clone_rocwmma
+        run: |
+          git clone https://github.com/rocm/rocwmma --branch rocm-6.2.4 --depth 1
+
       - name: ccache
         uses: hendrikmuhs/ccache-action@v1.2.16
         with:
@@ -1273,8 +1323,10 @@ jobs:
           cmake -G "Unix Makefiles" -B build -S . `
             -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" `
             -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" `
+            -DCMAKE_CXX_FLAGS="-I$($PWD.Path.Replace('\', '/'))/rocwmma/library/include/" `
             -DCMAKE_BUILD_TYPE=Release `
             -DAMDGPU_TARGETS=${{ matrix.gpu_target }} `
+            -DGGML_HIP_ROCWMMA_FATTN=ON `
             -DGGML_HIP=ON `
             -DGGML_RPC=ON
           cmake --build build -j ${env:NUMBER_OF_PROCESSORS}
@@ -1313,6 +1365,8 @@ jobs:
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
 
       - name: Build
         id: cmake_build
@@ -1328,15 +1382,40 @@ jobs:
             -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 \
             -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
           cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
-          sudo cmake --install build --config Release
 
       - name: xcodebuild for swift package
         id: xcodebuild
         run: |
-          xcodebuild -scheme llama-Package -destination 'generic/platform=iOS'
+          ./build-xcframework.sh
 
       - name: Build Xcode project
-        run: xcodebuild -project examples/llama.swiftui/llama.swiftui.xcodeproj -scheme llama.swiftui -sdk iphoneos CODE_SIGNING_REQUIRED=NO CODE_SIGN_IDENTITY= -destination 'generic/platform=iOS' build
+        run: xcodebuild -project examples/llama.swiftui/llama.swiftui.xcodeproj -scheme llama.swiftui -sdk iphoneos CODE_SIGNING_REQUIRED=NO CODE_SIGN_IDENTITY= -destination 'generic/platform=iOS' FRAMEWORK_FOLDER_PATH=./build-ios build
+
+      - name: Determine tag name
+        id: tag
+        shell: bash
+        run: |
+          BUILD_NUMBER="$(git rev-list --count HEAD)"
+          SHORT_HASH="$(git rev-parse --short=7 HEAD)"
+          if [[ "${{ env.BRANCH_NAME }}" == "master" ]]; then
+            echo "name=b${BUILD_NUMBER}" >> $GITHUB_OUTPUT
+          else
+            SAFE_NAME=$(echo "${{ env.BRANCH_NAME }}" | tr '/' '-')
+            echo "name=${SAFE_NAME}-b${BUILD_NUMBER}-${SHORT_HASH}" >> $GITHUB_OUTPUT
+          fi
+
+      - name: Pack artifacts
+        id: pack_artifacts
+        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        run: |
+          zip --symlinks -r llama-${{ steps.tag.outputs.name }}-xcframework.zip build-apple/llama.xcframework
+
+      - name: Upload artifacts
+        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        uses: actions/upload-artifact@v4
+        with:
+          path: llama-${{ steps.tag.outputs.name }}-xcframework.zip
+          name: llama-${{ steps.tag.outputs.name }}-xcframework
 
   android-build:
     runs-on: ubuntu-latest

package/src/llama.cpp/.github/workflows/server.yml

@@ -161,6 +161,8 @@ jobs:
       - name: Tests
         id: server_integration_tests
         if: ${{ matrix.sanitizer == '' }}
+        env:
+          GITHUB_ACTIONS: "true"
         run: |
           cd examples/server/tests
           ./tests.sh

package/src/llama.cpp/CMakeLists.txt

@@ -29,6 +29,8 @@ else()
     set(LLAMA_STANDALONE OFF)
 endif()
 
+option(LLAMA_USE_SYSTEM_GGML "Use system libggml" OFF)
+
 if (EMSCRIPTEN)
     set(BUILD_SHARED_LIBS_DEFAULT OFF)
 
@@ -145,7 +147,13 @@ endif()
 # 3rd-party
 #
 
-if (NOT TARGET ggml)
+if (LLAMA_USE_SYSTEM_GGML)
+    message(STATUS "Using system-provided libggml, skipping ggml build")
+    find_package(ggml REQUIRED)
+    add_library(ggml ALIAS ggml::ggml)
+endif()
+
+if (NOT TARGET ggml AND NOT LLAMA_USE_SYSTEM_GGML)
     add_subdirectory(ggml)
     # ... otherwise assume ggml is added by a parent CMakeLists.txt
 endif()

package/src/llama.cpp/cmake/common.cmake

@@ -1,3 +1,5 @@
+include("ggml/cmake/common.cmake")
+
 function(llama_add_compile_flags)
     if (LLAMA_FATAL_WARNINGS)
         if (CMAKE_CXX_COMPILER_ID MATCHES "GNU" OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")

package/src/llama.cpp/common/CMakeLists.txt

@@ -57,8 +57,7 @@ add_library(${TARGET} STATIC
     arg.h
     base64.hpp
     chat.cpp
-    chat.hpp
-    chat-template.hpp
+    chat.h
     common.cpp
     common.h
     console.cpp
@@ -68,7 +67,8 @@ add_library(${TARGET} STATIC
     llguidance.cpp
     log.cpp
     log.h
-    minja.hpp
+    minja/chat-template.hpp
+    minja/minja.hpp
     ngram-cache.cpp
     ngram-cache.h
     sampling.cpp