@fugood/llama.node 1.4.14 → 1.5.0-rc.0

package/lib/binding.ts

@@ -374,24 +374,20 @@ export type ModelInfo = {
   is_hybrid: boolean
   chatTemplates: {
     llamaChat: boolean
-    minja: {
+    jinja: {
       default: boolean
       defaultCaps: {
         tools: boolean
         toolCalls: boolean
-        toolResponses: boolean
         systemRole: boolean
         parallelToolCalls: boolean
-        toolCallId: boolean
       }
       toolUse: boolean
       toolUseCaps?: {
         tools: boolean
         toolCalls: boolean
-        toolResponses: boolean
         systemRole: boolean
         parallelToolCalls: boolean
-        toolCallId: boolean
       }
     }
   }
@@ -515,9 +511,20 @@ export interface LlamaContext {
   /**
    * Initialize multimodal support with a mmproj file
    * @param options Object containing path and optional use_gpu flag
+   * @param options.path Path to the multimodal projector model file (mmproj)
+   * @param options.use_gpu Whether to use GPU for multimodal processing (default: true)
+   * @param options.image_min_tokens Minimum number of tokens for image input (for dynamic resolution models)
+   * @param options.image_max_tokens Maximum number of tokens for image input (for dynamic resolution models).
+   *                                  Lower values reduce memory usage and improve speed for high-resolution images.
+   *                                  Recommended: 256-512 for faster inference, up to 4096 for maximum detail.
    * @returns boolean indicating if initialization was successful
    */
-  initMultimodal(options: { path: string; use_gpu?: boolean }): boolean
+  initMultimodal(options: {
+    path: string
+    use_gpu?: boolean
+    image_min_tokens?: number
+    image_max_tokens?: number
+  }): boolean
 
   /**
    * Check if multimodal support is enabled

package/lib/index.js

@@ -80,8 +80,8 @@ class LlamaContextWrapper {
         return this.ctx.getUsedDevices();
     }
     isJinjaSupported() {
-        const { minja } = this.ctx.getModelInfo().chatTemplates;
-        return !!(minja === null || minja === void 0 ? void 0 : minja.toolUse) || !!(minja === null || minja === void 0 ? void 0 : minja.default);
+        const { jinja } = this.ctx.getModelInfo().chatTemplates;
+        return !!(jinja === null || jinja === void 0 ? void 0 : jinja.toolUse) || !!(jinja === null || jinja === void 0 ? void 0 : jinja.default);
     }
     isLlamaChatSupported() {
         return !!this.ctx.getModelInfo().chatTemplates.llamaChat;

package/lib/index.ts

@@ -100,8 +100,8 @@ class LlamaContextWrapper {
   }
 
   isJinjaSupported(): boolean {
-    const { minja } = this.ctx.getModelInfo().chatTemplates
-    return !!minja?.toolUse || !!minja?.default
+    const { jinja } = this.ctx.getModelInfo().chatTemplates
+    return !!jinja?.toolUse || !!jinja?.default
   }
 
   isLlamaChatSupported(): boolean {
@@ -254,7 +254,12 @@ class LlamaContextWrapper {
     return this.ctx.getLoadedLoraAdapters()
   }
 
-  initMultimodal(options: { path: string; use_gpu?: boolean }): boolean {
+  initMultimodal(options: {
+    path: string
+    use_gpu?: boolean
+    image_min_tokens?: number
+    image_max_tokens?: number
+  }): boolean {
     return this.ctx.initMultimodal(options)
   }
 

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@fugood/llama.node",
   "access": "public",
-  "version": "1.4.14",
+  "version": "1.5.0-rc.0",
   "description": "An another Node binding of llama.cpp",
   "main": "lib/index.js",
   "scripts": {
@@ -72,20 +72,20 @@
     "CMakeLists.txt"
   ],
   "optionalDependencies": {
-    "@fugood/node-llama-darwin-arm64": "1.4.14",
-    "@fugood/node-llama-darwin-x64": "1.4.14",
-    "@fugood/node-llama-linux-arm64": "1.4.14",
-    "@fugood/node-llama-linux-arm64-cuda": "1.4.14",
-    "@fugood/node-llama-linux-arm64-snapdragon": "1.4.14",
-    "@fugood/node-llama-linux-arm64-vulkan": "1.4.14",
-    "@fugood/node-llama-linux-x64": "1.4.14",
-    "@fugood/node-llama-linux-x64-cuda": "1.4.14",
-    "@fugood/node-llama-linux-x64-vulkan": "1.4.14",
-    "@fugood/node-llama-win32-arm64": "1.4.14",
-    "@fugood/node-llama-win32-arm64-vulkan": "1.4.14",
-    "@fugood/node-llama-win32-x64": "1.4.14",
-    "@fugood/node-llama-win32-x64-cuda": "1.4.14",
-    "@fugood/node-llama-win32-x64-vulkan": "1.4.14"
+    "@fugood/node-llama-darwin-arm64": "1.5.0-rc.0",
+    "@fugood/node-llama-darwin-x64": "1.5.0-rc.0",
+    "@fugood/node-llama-linux-arm64": "1.5.0-rc.0",
+    "@fugood/node-llama-linux-arm64-cuda": "1.5.0-rc.0",
+    "@fugood/node-llama-linux-arm64-snapdragon": "1.5.0-rc.0",
+    "@fugood/node-llama-linux-arm64-vulkan": "1.5.0-rc.0",
+    "@fugood/node-llama-linux-x64": "1.5.0-rc.0",
+    "@fugood/node-llama-linux-x64-cuda": "1.5.0-rc.0",
+    "@fugood/node-llama-linux-x64-vulkan": "1.5.0-rc.0",
+    "@fugood/node-llama-win32-arm64": "1.5.0-rc.0",
+    "@fugood/node-llama-win32-arm64-vulkan": "1.5.0-rc.0",
+    "@fugood/node-llama-win32-x64": "1.5.0-rc.0",
+    "@fugood/node-llama-win32-x64-cuda": "1.5.0-rc.0",
+    "@fugood/node-llama-win32-x64-vulkan": "1.5.0-rc.0"
   },
   "devDependencies": {
     "@babel/preset-env": "^7.24.4",

package/scripts/llama.cpp.patch

@@ -1,8 +1,8 @@
 diff --git a/src/llama.cpp/common/CMakeLists.txt b/src/llama.cpp/common/CMakeLists.txt
-index f7b99159e..fa37fed19 100644
+index ae02c0bd7..f74d8bb26 100644
 --- a/src/llama.cpp/common/CMakeLists.txt
 +++ b/src/llama.cpp/common/CMakeLists.txt
-@@ -154,8 +154,14 @@ if (LLAMA_LLGUIDANCE)
+@@ -158,4 +158,11 @@ if (LLAMA_LLGUIDANCE)
      set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} llguidance ${LLGUIDANCE_PLATFORM_LIBS})
  endif ()
  
@@ -13,11 +13,8 @@ index f7b99159e..fa37fed19 100644
 +else()
 +    set(LLAMA_COMMON_WIN_LIBS "")
 +endif()
- 
++
 +target_link_libraries(${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} ${LLAMA_COMMON_WIN_LIBS} PUBLIC llama Threads::Threads)
- 
- #
- # copy the license files
 diff --git a/src/llama.cpp/common/chat-peg-parser.cpp b/src/llama.cpp/common/chat-peg-parser.cpp
 index 1bcba9cd8..b7cd68734 100644
 --- a/src/llama.cpp/common/chat-peg-parser.cpp
@@ -32,74 +29,89 @@ index 1bcba9cd8..b7cd68734 100644
  static std::string_view trim_trailing_space(std::string_view sv, int max = -1) {
      int count = 0;
 diff --git a/src/llama.cpp/common/chat.cpp b/src/llama.cpp/common/chat.cpp
-index 22e527bab..c3d0affca 100644
+index b29544dac..5fa2c6c17 100644
 --- a/src/llama.cpp/common/chat.cpp
 +++ b/src/llama.cpp/common/chat.cpp
-@@ -7,9 +7,6 @@
- #include "log.h"
- #include "regex-partial.h"
- 
--#include <minja/chat-template.hpp>
--#include <minja/minja.hpp>
--
- #include <algorithm>
- #include <cstdio>
- #include <cctype>
-@@ -135,16 +132,6 @@ std::vector<common_chat_msg_diff> common_chat_msg_diff::compute_diffs(const comm
-     return diffs;
+@@ -615,6 +615,37 @@ std::string common_chat_templates_source(const struct common_chat_templates * tm
+     return tmpls->template_default->source();
  }
  
--typedef minja::chat_template common_chat_template;
--
--struct common_chat_templates {
--    bool add_bos;
--    bool add_eos;
--    bool has_explicit_template; // Model had builtin template or template overridde was specified.
--    std::unique_ptr<common_chat_template> template_default; // always set (defaults to chatml)
--    std::unique_ptr<common_chat_template> template_tool_use;
--};
--
- struct templates_params {
-     json messages;
-     json tools;
-@@ -752,7 +739,7 @@ static std::string apply(
-         tmpl_inputs.extra_context.merge_patch(*additional_context);
++common_chat_template_caps common_chat_templates_get_caps(const struct common_chat_templates * tmpls, const std::string & variant) {
++    common_chat_template_caps result;
++    const common_chat_template * tmpl = nullptr;
++
++    if (!variant.empty() && variant == "tool_use") {
++        tmpl = tmpls->template_tool_use.get();
++    } else {
++        tmpl = tmpls->template_default.get();
++    }
++
++    if (tmpl) {
++        auto caps = tmpl->original_caps();
++        result.supports_tools = caps.supports_tools;
++        result.supports_tool_calls = caps.supports_tool_calls;
++        result.supports_system_role = caps.supports_system_role;
++        result.supports_parallel_tool_calls = caps.supports_parallel_tool_calls;
++    }
++
++    return result;
++}
++
++bool common_chat_templates_has_variant(const struct common_chat_templates * tmpls, const std::string & variant) {
++    if (variant.empty() || variant == "default") {
++        return tmpls->template_default != nullptr;
++    }
++    if (variant == "tool_use") {
++        return tmpls->template_tool_use != nullptr;
++    }
++    return false;
++}
++
+ common_chat_templates_ptr common_chat_templates_init(
+     const struct llama_model * model,
+     const std::string & chat_template_override,
+@@ -831,8 +862,9 @@ static std::string apply(
+     if (inputs.add_generation_prompt) {
+         inp["add_generation_prompt"] = true;
+     }
+-    if (inp["tools"].is_null()) {
+-        inp["tools"] = json::array();
++    // Remove tools key when null, so templates can check "{% if tools is defined %}"
++    if (inp["tools"].is_null() || (inp["tools"].is_array() && inp["tools"].empty())) {
++        inp.erase("tools");
      }
-     // TODO: add flag to control date/time, if only for testing purposes.
--    // tmpl_inputs.now = std::chrono::system_clock::now();
-+    tmpl_inputs.now = inputs.now;
  
-     minja::chat_template_options tmpl_opts;
-     // To avoid double BOS / EOS tokens, we're manually removing begining / trailing tokens
+     jinja::global_from_json(ctx, inp, inputs.mark_input);
 diff --git a/src/llama.cpp/common/chat.h b/src/llama.cpp/common/chat.h
-index 8bd4a325f..333b3301f 100644
+index 148801738..0317f1ab1 100644
 --- a/src/llama.cpp/common/chat.h
 +++ b/src/llama.cpp/common/chat.h
-@@ -10,7 +10,18 @@
- #include <vector>
- #include <map>
+@@ -222,6 +222,20 @@ common_chat_tool_choice common_chat_tool_choice_parse_oaicompat(const std::strin
+ 
+ bool common_chat_templates_support_enable_thinking(const common_chat_templates * chat_templates);
  
--struct common_chat_templates;
-+#include "minja/chat-template.hpp"
-+#include "minja/minja.hpp"
++// Template capabilities structure (for exposing capabilities to external code)
++struct common_chat_template_caps {
++    bool supports_tools = true;
++    bool supports_tool_calls = true;
++    bool supports_system_role = true;
++    bool supports_parallel_tool_calls = true;
++};
 +
-+typedef minja::chat_template common_chat_template;
++// Get template capabilities for a specific variant ("" for default, "tool_use" for tool_use template)
++common_chat_template_caps common_chat_templates_get_caps(const struct common_chat_templates * tmpls, const std::string & variant = "");
 +
-+struct common_chat_templates {
-+    bool add_bos;
-+    bool add_eos;
-+    bool has_explicit_template; // Model had builtin template or template overridde was specified.
-+    std::unique_ptr<common_chat_template> template_default; // always set (defaults to chatml)
-+    std::unique_ptr<common_chat_template> template_tool_use;
-+};
- 
- struct common_chat_tool_call {
-     std::string name;
++// Check if a template variant exists
++bool common_chat_templates_has_variant(const struct common_chat_templates * tmpls, const std::string & variant);
++
+ // Parses a JSON array of messages in OpenAI's chat completion API format.
+ // T can be std::string containing JSON or nlohmann::ordered_json
+ template <class T> std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const T & messages);
 diff --git a/src/llama.cpp/common/common.cpp b/src/llama.cpp/common/common.cpp
-index 744f0b4ee..04fcebb9e 100644
+index 26250abb6..72ceddcc7 100644
 --- a/src/llama.cpp/common/common.cpp
 +++ b/src/llama.cpp/common/common.cpp
-@@ -1361,6 +1361,7 @@ struct llama_model_params common_model_params_to_llama(common_params & params) {
+@@ -1360,6 +1360,7 @@ struct llama_model_params common_model_params_to_llama(common_params & params) {
          mparams.devices = params.devices.data();
      }
  
@@ -108,10 +120,10 @@ index 744f0b4ee..04fcebb9e 100644
      mparams.main_gpu        = params.main_gpu;
      mparams.split_mode      = params.split_mode;
 diff --git a/src/llama.cpp/common/common.h b/src/llama.cpp/common/common.h
-index 7794c0268..5b77ae0c3 100644
+index b9566df62..c9425ad2f 100644
 --- a/src/llama.cpp/common/common.h
 +++ b/src/llama.cpp/common/common.h
-@@ -310,6 +310,7 @@ struct lr_opt {
+@@ -314,6 +314,7 @@ struct lr_opt {
  struct ggml_opt_optimizer_params common_opt_lr_pars(void * userdata);
  
  struct common_params {
@@ -133,10 +145,10 @@ index 7622d0bf4..d2edcfddb 100644
              check_cxx_compiler_flag(-mfp16-format=ieee GGML_COMPILER_SUPPORTS_FP16_FORMAT_I3E)
              if (NOT "${GGML_COMPILER_SUPPORTS_FP16_FORMAT_I3E}" STREQUAL "")
 diff --git a/src/llama.cpp/ggml/src/ggml-hexagon/ggml-hexagon.cpp b/src/llama.cpp/ggml/src/ggml-hexagon/ggml-hexagon.cpp
-index 365a24b49..83bf4ee62 100644
+index 5b835c11c..681c00504 100644
 --- a/src/llama.cpp/ggml/src/ggml-hexagon/ggml-hexagon.cpp
 +++ b/src/llama.cpp/ggml/src/ggml-hexagon/ggml-hexagon.cpp
-@@ -2798,9 +2798,24 @@ static const char * ggml_backend_hexagon_device_get_description(ggml_backend_dev
+@@ -2819,9 +2819,24 @@ static const char * ggml_backend_hexagon_device_get_description(ggml_backend_dev
      GGML_UNUSED(dev);
  }
  
@@ -163,7 +175,7 @@ index 365a24b49..83bf4ee62 100644
      *total = *free;
  
      GGML_UNUSED(dev);
-@@ -3010,10 +3025,17 @@ ggml_hexagon_registry::ggml_hexagon_registry(ggml_backend_reg_t reg) {
+@@ -3056,10 +3071,17 @@ ggml_hexagon_registry::ggml_hexagon_registry(ggml_backend_reg_t reg) {
          }
      }
  
@@ -185,7 +197,7 @@ index 365a24b49..83bf4ee62 100644
  
      GGML_LOG_INFO("ggml-hex: Hexagon Arch version v%d\n", opt_arch);
  
-@@ -3026,6 +3048,8 @@ ggml_hexagon_registry::ggml_hexagon_registry(ggml_backend_reg_t reg) {
+@@ -3072,6 +3094,8 @@ ggml_hexagon_registry::ggml_hexagon_registry(ggml_backend_reg_t reg) {
          } catch (const std::exception & exc) {
              GGML_LOG_ERROR("ggml-hex: failed to create device/session %zu\n", i);
              devices[i].context = nullptr;

package/src/LlamaContext.cpp

@@ -10,6 +10,7 @@
 #include "DecodeAudioTokenWorker.h"
 #include "ggml.h"
 #include "gguf.h"
+#include "chat.h"
 #include "json-schema-to-grammar.h"
 #include <nlohmann/json.hpp>
 #include "llama-impl.h"
@@ -600,40 +601,27 @@ Napi::Value LlamaContext::GetModelInfo(const Napi::CallbackInfo &info) {
 
   Napi::Object chatTemplates = Napi::Object::New(info.Env());
   chatTemplates.Set("llamaChat", _rn_ctx->validateModelChatTemplate(false, nullptr));
-  Napi::Object minja = Napi::Object::New(info.Env());
-  minja.Set("default", _rn_ctx->validateModelChatTemplate(true, nullptr));
+  Napi::Object jinja = Napi::Object::New(info.Env());
+  jinja.Set("default", _rn_ctx->validateModelChatTemplate(true, nullptr));
   Napi::Object defaultCaps = Napi::Object::New(info.Env());
-  auto default_tmpl = _rn_ctx->templates.get()->template_default.get();
-  auto default_tmpl_caps = default_tmpl->original_caps();
-  defaultCaps.Set(
-      "tools",
-      default_tmpl_caps.supports_tools);
-  defaultCaps.Set(
-      "toolCalls",
-      default_tmpl_caps.supports_tool_calls);
-  defaultCaps.Set("toolResponses", default_tmpl_caps.supports_tool_responses);
-  defaultCaps.Set(
-      "systemRole",
-      default_tmpl_caps.supports_system_role);
-  defaultCaps.Set("parallelToolCalls", default_tmpl_caps.supports_parallel_tool_calls);
-  defaultCaps.Set("toolCallId", default_tmpl_caps.supports_tool_call_id);
-  minja.Set("defaultCaps", defaultCaps);
-  minja.Set("toolUse", _rn_ctx->validateModelChatTemplate(true, "tool_use"));
-  if (_rn_ctx->validateModelChatTemplate(true, "tool_use")) {
+  auto default_caps = common_chat_templates_get_caps(_rn_ctx->templates.get(), "");
+  defaultCaps.Set("tools", default_caps.supports_tools);
+  defaultCaps.Set("toolCalls", default_caps.supports_tool_calls);
+  defaultCaps.Set("systemRole", default_caps.supports_system_role);
+  defaultCaps.Set("parallelToolCalls", default_caps.supports_parallel_tool_calls);
+  jinja.Set("defaultCaps", defaultCaps);
+  bool hasToolUse = common_chat_templates_has_variant(_rn_ctx->templates.get(), "tool_use");
+  jinja.Set("toolUse", hasToolUse);
+  if (hasToolUse) {
     Napi::Object toolUseCaps = Napi::Object::New(info.Env());
-    auto tool_use_tmpl = _rn_ctx->templates.get()->template_tool_use.get();
-    auto tool_use_tmpl_caps = tool_use_tmpl->original_caps();
-    toolUseCaps.Set(
-        "tools",
-        tool_use_tmpl_caps.supports_tools);
-    toolUseCaps.Set("toolCalls", tool_use_tmpl_caps.supports_tool_calls);
-    toolUseCaps.Set("toolResponses", tool_use_tmpl_caps.supports_tool_responses);
-    toolUseCaps.Set("systemRole", tool_use_tmpl_caps.supports_system_role);
-    toolUseCaps.Set("parallelToolCalls", tool_use_tmpl_caps.supports_parallel_tool_calls);
-    toolUseCaps.Set("toolCallId", tool_use_tmpl_caps.supports_tool_call_id);
-    minja.Set("toolUseCaps", toolUseCaps);
-  }
-  chatTemplates.Set("minja", minja);
+    auto tool_use_caps = common_chat_templates_get_caps(_rn_ctx->templates.get(), "tool_use");
+    toolUseCaps.Set("tools", tool_use_caps.supports_tools);
+    toolUseCaps.Set("toolCalls", tool_use_caps.supports_tool_calls);
+    toolUseCaps.Set("systemRole", tool_use_caps.supports_system_role);
+    toolUseCaps.Set("parallelToolCalls", tool_use_caps.supports_parallel_tool_calls);
+    jinja.Set("toolUseCaps", toolUseCaps);
+  }
+  chatTemplates.Set("jinja", jinja);
   details.Set("chatTemplates", chatTemplates);
 
   details.Set("metadata", metadata);
@@ -1333,7 +1321,7 @@ extern "C" void cleanup_logging() {
 }
 
 
-// initMultimodal(options: { path: string, use_gpu?: boolean }): boolean
+// initMultimodal(options: { path: string, use_gpu?: boolean, image_min_tokens?: number, image_max_tokens?: number }): boolean
 Napi::Value LlamaContext::InitMultimodal(const Napi::CallbackInfo &info) {
   Napi::Env env = info.Env();
 
@@ -1345,6 +1333,15 @@ Napi::Value LlamaContext::InitMultimodal(const Napi::CallbackInfo &info) {
   auto options = info[0].As<Napi::Object>();
   auto mmproj_path = options.Get("path").ToString().Utf8Value();
   auto use_gpu = options.Get("use_gpu").ToBoolean().Value();
+  int image_min_tokens = -1;
+  int image_max_tokens = -1;
+
+  if (options.Has("image_min_tokens") && options.Get("image_min_tokens").IsNumber()) {
+    image_min_tokens = options.Get("image_min_tokens").ToNumber().Int32Value();
+  }
+  if (options.Has("image_max_tokens") && options.Get("image_max_tokens").IsNumber()) {
+    image_max_tokens = options.Get("image_max_tokens").ToNumber().Int32Value();
+  }
 
   if (mmproj_path.empty()) {
     Napi::TypeError::New(env, "mmproj path is required")
@@ -1360,7 +1357,7 @@ Napi::Value LlamaContext::InitMultimodal(const Napi::CallbackInfo &info) {
 
   // Disable ctx_shift before initializing multimodal
   _rn_ctx->params.ctx_shift = false;
-  bool result = _rn_ctx->initMultimodal(mmproj_path, use_gpu);
+  bool result = _rn_ctx->initMultimodal(mmproj_path, use_gpu, image_min_tokens, image_max_tokens);
   if (!result) {
     Napi::Error::New(env, "Failed to initialize multimodal context")
         .ThrowAsJavaScriptException();

package/src/llama.cpp/CMakeLists.txt

@@ -111,11 +111,16 @@ option(LLAMA_BUILD_SERVER   "llama: build server example" ${LLAMA_STANDALONE})
 option(LLAMA_TOOLS_INSTALL  "llama: install tools"        ${LLAMA_TOOLS_INSTALL_DEFAULT})
 
 # 3rd party libs
-option(LLAMA_CURL       "llama: use libcurl to download model from an URL" ON)
-option(LLAMA_HTTPLIB    "llama: if libcurl is disabled, use httplib to download model from an URL" ON)
-option(LLAMA_OPENSSL    "llama: use openssl to support HTTPS" OFF)
+option(LLAMA_HTTPLIB    "llama: httplib for downloading functionality" ON)
+option(LLAMA_OPENSSL    "llama: use openssl to support HTTPS" ON)
 option(LLAMA_LLGUIDANCE "llama-common: include LLGuidance library for structured output in common utils" OFF)
 
+# deprecated
+option(LLAMA_CURL "llama: use libcurl to download model from an URL" OFF)
+if (LLAMA_CURL)
+    message(WARNING "LLAMA_CURL option is deprecated and will be ignored")
+endif()
+
 # Required for relocatable CMake package
 include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/build-info.cmake)
 include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/common.cmake)
@@ -182,6 +187,9 @@ if (NOT MSVC)
     endif()
 endif()
 
+include("cmake/license.cmake")
+license_add_file("llama.cpp" "LICENSE")
+
 #
 # 3rd-party
 #
@@ -209,11 +217,6 @@ add_subdirectory(src)
 # utils, programs, examples and tests
 #
 
-if (NOT LLAMA_BUILD_COMMON)
-    message(STATUS "LLAMA_BUILD_COMMON is OFF, disabling LLAMA_CURL")
-    set(LLAMA_CURL OFF)
-endif()
-
 if (LLAMA_BUILD_COMMON)
     add_subdirectory(common)
     if (LLAMA_HTTPLIB)
@@ -235,6 +238,19 @@ if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TOOLS)
     add_subdirectory(tools)
 endif()
 
+# Automatically add all files from the 'licenses' directory
+file(GLOB EXTRA_LICENSES "${CMAKE_SOURCE_DIR}/licenses/LICENSE-*")
+
+foreach(FILE_PATH ${EXTRA_LICENSES})
+    get_filename_component(FILE_NAME "${FILE_PATH}" NAME)
+    string(REGEX REPLACE "^LICENSE-" "" NAME "${FILE_NAME}")
+    license_add_file("${NAME}" "${FILE_PATH}")
+endforeach()
+
+if (LLAMA_BUILD_COMMON)
+    license_generate(common)
+endif()
+
 #
 # install
 #

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

@@ -60,6 +60,8 @@ add_library(${TARGET} STATIC
     common.h
     console.cpp
     console.h
+    debug.cpp
+    debug.h
     download.cpp
     download.h
     http.h
@@ -83,6 +85,18 @@ add_library(${TARGET} STATIC
     speculative.h
     unicode.cpp
     unicode.h
+    jinja/lexer.cpp
+    jinja/lexer.h
+    jinja/parser.cpp
+    jinja/parser.h
+    jinja/runtime.cpp
+    jinja/runtime.h
+    jinja/value.cpp
+    jinja/value.h
+    jinja/string.cpp
+    jinja/string.h
+    jinja/caps.cpp
+    jinja/caps.h
     )
 
 target_include_directories(${TARGET} PUBLIC . ../vendor)
@@ -95,17 +109,7 @@ endif()
 # TODO: use list(APPEND LLAMA_COMMON_EXTRA_LIBS ...)
 set(LLAMA_COMMON_EXTRA_LIBS build_info)
 
-if (LLAMA_CURL)
-    # Use curl to download model url
-    find_package(CURL)
-    if (NOT CURL_FOUND)
-        message(FATAL_ERROR "Could NOT find CURL. Hint: to disable this feature, set -DLLAMA_CURL=OFF")
-    endif()
-    target_compile_definitions(${TARGET} PUBLIC LLAMA_USE_CURL)
-    include_directories(${CURL_INCLUDE_DIRS})
-    set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} ${CURL_LIBRARIES})
-elseif (LLAMA_HTTPLIB)
-    # otherwise, use cpp-httplib
+if (LLAMA_HTTPLIB)
     target_compile_definitions(${TARGET} PUBLIC LLAMA_USE_HTTPLIB)
     set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} cpp-httplib)
 endif()
@@ -162,26 +166,3 @@ else()
 endif()
 
 target_link_libraries(${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} ${LLAMA_COMMON_WIN_LIBS} PUBLIC llama Threads::Threads)
-
-#
-# copy the license files
-#
-
-# Check if running in GitHub Actions
-if (DEFINED ENV{GITHUB_ACTIONS} AND "$ENV{GITHUB_ACTIONS}" STREQUAL "true")
-    message(STATUS "Running inside GitHub Actions - copying license files")
-
-    # Copy all files from licenses/ to build/bin/
-    file(GLOB LICENSE_FILES "${CMAKE_SOURCE_DIR}/licenses/*")
-    foreach(LICENSE_FILE ${LICENSE_FILES})
-        get_filename_component(FILENAME ${LICENSE_FILE} NAME)
-        add_custom_command(
-            POST_BUILD
-            TARGET ${TARGET}
-            COMMAND ${CMAKE_COMMAND} -E copy_if_different
-                "${LICENSE_FILE}"
-                "$<TARGET_FILE_DIR:llama>/${FILENAME}"
-            COMMENT "Copying ${FILENAME} to ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}")
-        message(STATUS "Copying ${LICENSE_FILE} to ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${FILENAME}")
-    endforeach()
-endif()