async memory streams

.github/CODEOWNERS

@@ -0,0 +1 @@
+* @tetherto/ai-runtime-bk-models

.github/workflows/approval-check-worker.yml

@@ -0,0 +1,19 @@
+name: Approval Check Worker
+
+on:
+  pull_request_review:
+    types: [submitted, dismissed]
+
+jobs:
+  check-approvals:
+    permissions:
+      contents: write
+      pull-requests: write
+      statuses: write
+      issues: write
+
+    uses: tetherto/qvac-devops/.github/workflows/approval-check-worker.yml@production-workflows-tag
+    secrets: inherit
+    with:
+      pr_number: ${{ github.event.pull_request.number }}
+      pr_sha: ${{ github.event.pull_request.head.sha }}

CMakeLists.txt

@@ -79,7 +79,10 @@ option(LLAMA_BUILD_COMMON "llama: build common utils library" ${LLAMA_STANDALONE
 option(LLAMA_BUILD_TESTS    "llama: build tests"          ${LLAMA_STANDALONE})
 option(LLAMA_BUILD_TOOLS    "llama: build tools"          ${LLAMA_STANDALONE})
 option(LLAMA_BUILD_EXAMPLES "llama: build examples"       ${LLAMA_STANDALONE})
-option(LLAMA_BUILD_SERVER   "llama: build server example" ${LLAMA_STANDALONE})
+option(LLAMA_BUILD_SERVER   "llama: build server example" OFF)
+
+# specific extras
+option(LLAMA_MTMD "llama: multimodal support" OFF)
 
 # 3rd party libs
 option(LLAMA_CURL       "llama: use libcurl to download model from an URL" ON)
@@ -192,6 +195,10 @@ if (LLAMA_BUILD_COMMON)
     add_subdirectory(common)
 endif()
 
+if(LLAMA_BUILD_EXAMPLES OR LLAMA_BUILD_TESTS)
+    add_subdirectory(common_test)
+endif()
+
 if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TESTS AND NOT CMAKE_JS_VERSION)
     include(CTest)
     add_subdirectory(tests)
@@ -206,14 +213,18 @@ if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TOOLS)
     add_subdirectory(tools)
 endif()
 
+if (LLAMA_MTMD)
+  add_subdirectory(tools/mtmd)
+endif()
+
 #
 # install
 #
 
 include(GNUInstallDirs)
 include(CMakePackageConfigHelpers)
 
-set(LLAMA_INCLUDE_INSTALL_DIR ${CMAKE_INSTALL_INCLUDEDIR} CACHE PATH "Location of header  files")
+set(LLAMA_INCLUDE_INSTALL_DIR ${CMAKE_INSTALL_INCLUDEDIR}/llama CACHE PATH "Location of header  files")
 set(LLAMA_LIB_INSTALL_DIR     ${CMAKE_INSTALL_LIBDIR}     CACHE PATH "Location of library files")
 set(LLAMA_BIN_INSTALL_DIR     ${CMAKE_INSTALL_BINDIR}     CACHE PATH "Location of binary  files")
 
@@ -225,15 +236,46 @@ set_target_properties(llama
     PROPERTIES
         PUBLIC_HEADER "${LLAMA_PUBLIC_HEADERS}")
 
-install(TARGETS llama LIBRARY PUBLIC_HEADER)
+install(
+  TARGETS llama
+  EXPORT llama-targets
+  PUBLIC_HEADER
+    DESTINATION ${LLAMA_INCLUDE_INSTALL_DIR})
+
+if (LLAMA_BUILD_COMMON)
+
+  install(
+    TARGETS common build_info
+    EXPORT llama-targets
+    PUBLIC_HEADER
+      DESTINATION ${LLAMA_INCLUDE_INSTALL_DIR}/common)
+
+endif()
+
+if (LLAMA_MTMD)
+
+  install(
+    TARGETS mtmd
+    EXPORT llama-targets
+    PUBLIC_HEADER
+      DESTINATION ${LLAMA_INCLUDE_INSTALL_DIR}/mtmd)
+
+endif()
+
+install(
+  EXPORT llama-targets
+  FILE llama-targets.cmake
+  NAMESPACE llama::
+  DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/llama)
+
+install(
+  FILES ${CMAKE_CURRENT_BINARY_DIR}/llama-config.cmake
+  DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/llama)
 
 configure_package_config_file(
-        ${CMAKE_CURRENT_SOURCE_DIR}/cmake/llama-config.cmake.in
-        ${CMAKE_CURRENT_BINARY_DIR}/llama-config.cmake
-    INSTALL_DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/llama
-    PATH_VARS LLAMA_INCLUDE_INSTALL_DIR
-              LLAMA_LIB_INSTALL_DIR
-              LLAMA_BIN_INSTALL_DIR )
+  ${CMAKE_CURRENT_SOURCE_DIR}/cmake/llama-config.cmake.in
+  ${CMAKE_CURRENT_BINARY_DIR}/llama-config.cmake
+  INSTALL_DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/llama)
 
 write_basic_package_version_file(
         ${CMAKE_CURRENT_BINARY_DIR}/llama-version.cmake
@@ -242,7 +284,7 @@ write_basic_package_version_file(
 
 install(FILES ${CMAKE_CURRENT_BINARY_DIR}/llama-config.cmake
               ${CMAKE_CURRENT_BINARY_DIR}/llama-version.cmake
-        DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/llama)
+        DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/llama)
 
 install(
     FILES convert_hf_to_gguf.py

cmake/llama-config.cmake.in

@@ -5,26 +5,9 @@ set(LLAMA_SHARED_LIB   @BUILD_SHARED_LIBS@)
 
 @PACKAGE_INIT@
 
-set_and_check(LLAMA_INCLUDE_DIR "@PACKAGE_LLAMA_INCLUDE_INSTALL_DIR@")
-set_and_check(LLAMA_LIB_DIR     "@PACKAGE_LLAMA_LIB_INSTALL_DIR@")
-set_and_check(LLAMA_BIN_DIR     "@PACKAGE_LLAMA_BIN_INSTALL_DIR@")
+include(CMakeFindDependencyMacro)
+find_dependency(ggml CONFIG REQUIRED)
 
-find_package(ggml REQUIRED HINTS ${LLAMA_LIB_DIR}/cmake)
+include("${CMAKE_CURRENT_LIST_DIR}/llama-targets.cmake")
 
-find_library(llama_LIBRARY llama
-    REQUIRED
-    HINTS ${LLAMA_LIB_DIR}
-    NO_CMAKE_FIND_ROOT_PATH
-)
-
-add_library(llama UNKNOWN IMPORTED)
-set_target_properties(llama
-    PROPERTIES
-        INTERFACE_INCLUDE_DIRECTORIES "${LLAMA_INCLUDE_DIR}"
-        INTERFACE_LINK_LIBRARIES "ggml::ggml;ggml::ggml-base;"
-        IMPORTED_LINK_INTERFACE_LANGUAGES "CXX"
-        IMPORTED_LOCATION "${llama_LIBRARY}"
-        INTERFACE_COMPILE_FEATURES c_std_90
-        POSITION_INDEPENDENT_CODE ON)
-
-check_required_components(Llama)
+check_required_components(llama)

common/CMakeLists.txt

@@ -44,34 +44,43 @@ endif()
 
 set(TARGET common)
 
+set(${TARGET}_HEADERS
+  arg.h
+  base64.hpp
+  chat-parser.h
+  chat.h
+  common.h
+  console.h
+  json-partial.h
+  json-schema-to-grammar.h
+  log.h
+  ngram-cache.h
+  regex-partial.h
+  sampling.h
+  speculative.h
+)
+
+list(TRANSFORM ${TARGET}_HEADERS PREPEND ${CMAKE_SOURCE_DIR}/common/)
+
 add_library(${TARGET} STATIC
     arg.cpp
-    arg.h
-    base64.hpp
     chat-parser.cpp
-    chat-parser.h
     chat.cpp
-    chat.h
     common.cpp
-    common.h
     console.cpp
-    console.h
     json-partial.cpp
-    json-partial.h
     json-schema-to-grammar.cpp
     llguidance.cpp
     log.cpp
-    log.h
     ngram-cache.cpp
-    ngram-cache.h
     regex-partial.cpp
-    regex-partial.h
     sampling.cpp
-    sampling.h
     speculative.cpp
-    speculative.h
+    ${${TARGET}_HEADERS}
     )
 
+set_target_properties(${TARGET} PROPERTIES PUBLIC_HEADER "${${TARGET}_HEADERS}")
+
 if (BUILD_SHARED_LIBS)
     set_target_properties(${TARGET} PROPERTIES POSITION_INDEPENDENT_CODE ON)
 endif()
@@ -133,7 +142,12 @@ if (LLAMA_LLGUIDANCE)
     set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} llguidance ${LLGUIDANCE_PLATFORM_LIBS})
 endif ()
 
-target_include_directories(${TARGET} PUBLIC . ../vendor)
+target_include_directories(
+  ${TARGET}
+  PUBLIC
+    $<BUILD_INTERFACE:${CMAKE_SOURCE_DIR}/common$<SEMICOLON>${CMAKE_SOURCE_DIR}/vendor>
+    $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}>)
+
 target_compile_features   (${TARGET} PUBLIC cxx_std_17)
 target_link_libraries     (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)
 

common/common.cpp

@@ -899,15 +899,7 @@ std::string fs_get_cache_file(const std::string & filename) {
 // Model utils
 //
 
-struct common_init_result common_init_from_params(common_params & params) {
-    common_init_result iparams;
-    auto mparams = common_model_params_to_llama(params);
-
-    llama_model * model = llama_model_load_from_file(params.model.path.c_str(), mparams);
-    if (model == NULL) {
-        LOG_ERR("%s: failed to load model '%s'\n", __func__, params.model.path.c_str());
-        return iparams;
-    }
+struct common_init_result common_init_from_model_and_params(llama_model* model, common_init_result iparams, common_params & params) {
 
     const llama_vocab * vocab = llama_model_get_vocab(model);
 
@@ -1074,6 +1066,19 @@ struct common_init_result common_init_from_params(common_params & params) {
     return iparams;
 }
 
+struct common_init_result common_init_from_params(common_params & params) {
+    common_init_result iparams;
+    auto               mparams = common_model_params_to_llama(params);
+
+    llama_model * model = llama_model_load_from_file(params.model.path.c_str(), mparams);
+    if (model == NULL) {
+        LOG_ERR("%s: failed to load model '%s'\n", __func__, params.model.path.c_str());
+        return iparams;
+    }
+
+    return common_init_from_model_and_params(model, std::move(iparams), params);
+}
+
 std::string get_model_endpoint() {
     const char * model_endpoint_env = getenv("MODEL_ENDPOINT");
     // We still respect the use of environment-variable "HF_ENDPOINT" for backward-compatibility.

common/common.h

@@ -567,6 +567,8 @@ struct common_init_result {
 };
 
 struct common_init_result     common_init_from_params(common_params & params);
+struct common_init_result     common_init_from_model_and_params(llama_model * model, common_init_result iparams,
+                                                                common_params & params);
 
 struct llama_model_params     common_model_params_to_llama  (      common_params & params);
 struct llama_context_params   common_context_params_to_llama(const common_params & params);

common_test/CMakeLists.txt

@@ -0,0 +1,15 @@
+# common_test library for load_into_memory.h and uint8-buff-stream.h
+
+set(TARGET llama-common-test)
+
+add_library(${TARGET} INTERFACE)
+
+target_include_directories(${TARGET} INTERFACE
+    ${CMAKE_CURRENT_SOURCE_DIR}
+)
+
+target_compile_definitions(${TARGET} INTERFACE LLAMA_COMMON_TEST_HEADERS)
+
+target_compile_features(${TARGET} INTERFACE cxx_std_17)
+
+target_link_libraries(${TARGET} INTERFACE common)