Merge pull request #571 from janhq/update-dev-from-master-2026-06-26-01-15

Sync master with upstream release b9802

Author: jan-service-account

.github/labeler.yml

@@ -35,8 +35,20 @@ AMD ZenDNN:
 documentation:
     - changed-files:
         - any-glob-to-any-file:
+            - "**/*.md"
             - docs/**
             - media/**
+examples:
+    - all:
+        - changed-files:
+            - any-glob-to-any-file:
+                - app/**
+                - examples/**
+                - tools/**
+            - all-globs-to-all-files:
+                - '!tools/server/**'
+                - '!tools/mtmd/**'
+                - '!tools/ui/**'
 testing:
     - changed-files:
         - any-glob-to-any-file:
@@ -47,28 +59,12 @@ build:
             - cmake/**
             - CMakeLists.txt
             - CMakePresets.json
-examples:
-    - changed-files:
-        - any-glob-to-any-file:
-            - examples/**
-            - tools/**
 devops:
     - changed-files:
         - any-glob-to-any-file:
             - .devops/**
             - .github/**
             - ci/**
-python:
-    - changed-files:
-        - any-glob-to-any-file:
-            - "**/*.py"
-            - requirements/**
-            - gguf-py/**
-            - .flake8
-script:
-    - changed-files:
-        - any-glob-to-any-file:
-            - scripts/**
 android:
     - changed-files:
         - any-glob-to-any-file:
@@ -81,9 +77,20 @@ server:
     - changed-files:
         - any-glob-to-any-file:
             - tools/server/**
-
-
-
+mtmd:
+    - changed-files:
+        - any-glob-to-any-file:
+            - tools/mtmd/**
+conversion:
+    - changed-files:
+        - any-glob-to-any-file:
+            - conversion/**
+            - convert_*.py
+            - gguf-py/**
+vendor:
+    - changed-files:
+        - any-glob-to-any-file:
+            - vendor/**
 ggml:
     - changed-files:
         - any-glob-to-any-file:

CMakeLists.txt

@@ -222,6 +222,16 @@ if (LLAMA_BUILD_APP)
     add_subdirectory(app)
 endif()
 
+# Standalone libmtmd build without pulling in the rest of the tools/ tree.
+# Useful when packaging just the mtmd library for language bindings (e.g. an
+# Apple XCFramework, or a WASM build). When the full tools build is enabled,
+# mtmd is already built by the tools/ subdirectory above; this hook only fires
+# when LLAMA_BUILD_TOOLS is OFF to avoid double-adding the target.
+option(LLAMA_BUILD_MTMD "llama: build tools/mtmd library standalone" OFF)
+if (LLAMA_BUILD_MTMD AND NOT (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TOOLS))
+    add_subdirectory(tools/mtmd)
+endif()
+
 #
 # install
 #

app/CMakeLists.txt

@@ -1,6 +1,6 @@
 set(TARGET llama-app)
 
-add_executable(${TARGET} llama.cpp)
+add_executable(${TARGET} llama.cpp download.cpp)
 set_target_properties(${TARGET} PROPERTIES OUTPUT_NAME llama)
 
 target_link_libraries(${TARGET} PRIVATE

app/download.cpp

@@ -0,0 +1,71 @@
+#include "arg.h"
+#include "common.h"
+#include "download.h"
+#include "log.h"
+
+#include <cstdio>
+#include <filesystem>
+
+static void print_usage(int /*argc*/, char ** argv) {
+    printf(
+        "\nexamples:\n"
+        "  %s -hf ggml-org/gemma-3-4b-it-qat-GGUF\n"
+        "  %s -hf ggml-org/gemma-3-4b-it-qat-GGUF:Q4_K_M\n"
+        "  %s -hf ggml-org/models -hff model.gguf\n"
+        "  %s -mu https://example.com/model.gguf -m model.gguf\n"
+        "\n",
+        argv[0], argv[0], argv[0], argv[0]
+    );
+}
+
+int llama_download(int argc, char ** argv);
+
+int llama_download(int argc, char ** argv) {
+    common_init();
+
+    common_params params;
+    params.verbosity = LOG_LEVEL_ERROR;
+
+    if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_DOWNLOAD, print_usage)) {
+        return 1;
+    }
+
+    const bool has_source = !params.model.hf_repo.empty() || !params.model.url.empty() ||
+                            !params.model.path.empty()    || !params.model.docker_repo.empty();
+    if (!has_source) {
+        fprintf(stderr, "error: no model source specified (use --hf-repo, --model-url, --model or --docker-repo)\n");
+        return 1;
+    }
+
+    try {
+        common_models_handler handler = common_models_handler_init(params, LLAMA_EXAMPLE_DOWNLOAD);
+        common_models_handler_apply(handler, params);
+    } catch (const std::exception & e) {
+        fprintf(stderr, "error: %s\n", e.what());
+        return 1;
+    }
+
+    if (!params.models_preset.empty()) {
+        // -hf pointed at a preset repo: print the preset path and stop
+        printf("%s\n", params.models_preset.c_str());
+        return 0;
+    }
+    if (params.model.path.empty()) {
+        fprintf(stderr, "error: model download failed\n");
+        return 1;
+    }
+    if (!std::filesystem::exists(params.model.path)) {
+        fprintf(stderr, "error: model file does not exist: %s\n", params.model.path.c_str());
+        return 1;
+    }
+
+    printf("%s\n", params.model.path.c_str());
+    if (!params.mmproj.path.empty()) {
+        printf("%s\n", params.mmproj.path.c_str());
+    }
+    if (!params.speculative.draft.mparams.path.empty()) {
+        printf("%s\n", params.speculative.draft.mparams.path.c_str());
+    }
+
+    return 0;
+}

app/llama.cpp

@@ -19,6 +19,7 @@ int llama_batched_bench(int argc, char ** argv);
 int llama_fit_params(int argc, char ** argv);
 int llama_quantize(int argc, char ** argv);
 int llama_perplexity(int argc, char ** argv);
+int llama_download(int argc, char ** argv);
 
 // Self-update is only supported for binaries built with llama-install.sh
 static int llama_update(int argc, char ** argv) {
@@ -61,6 +62,7 @@ static const command cmds[] = {
     {"serve",         "HTTP API server",                                    {"server"},   false,         llama_server       },
     {"cli",           "Command-line interactive interface",                 {"client"},   false,         llama_cli          },
     {"update",        "Update llama to the latest release",                 {},           UPDATE_HIDDEN, llama_update       },
+    {"download",      "Download a model",                                   {"get"},      false,         llama_download     },
     {"completion",    "Text completion",                                    {"complete"}, true,          llama_completion   },
     {"bench",         "Benchmark prompt processing and text generation",    {},           true,          llama_bench        },
     {"batched-bench", "Benchmark batched decoding performance",             {},           true,          llama_batched_bench},

build-xcframework.sh

@@ -13,6 +13,7 @@ LLAMA_BUILD_EXAMPLES=OFF
 LLAMA_BUILD_TOOLS=OFF
 LLAMA_BUILD_TESTS=OFF
 LLAMA_BUILD_SERVER=OFF
+LLAMA_BUILD_MTMD=ON
 GGML_METAL=ON
 GGML_METAL_EMBED_LIBRARY=ON
 GGML_BLAS_DEFAULT=ON
@@ -39,6 +40,7 @@ COMMON_CMAKE_ARGS=(
     -DLLAMA_BUILD_TOOLS=${LLAMA_BUILD_TOOLS}
     -DLLAMA_BUILD_TESTS=${LLAMA_BUILD_TESTS}
     -DLLAMA_BUILD_SERVER=${LLAMA_BUILD_SERVER}
+    -DLLAMA_BUILD_MTMD=${LLAMA_BUILD_MTMD}
     -DGGML_METAL_EMBED_LIBRARY=${GGML_METAL_EMBED_LIBRARY}
     -DGGML_BLAS_DEFAULT=${GGML_BLAS_DEFAULT}
     -DGGML_METAL=${GGML_METAL}
@@ -126,6 +128,8 @@ setup_framework_structure() {
     cp ggml/include/ggml-cpu.h     ${header_path}
     cp ggml/include/ggml-blas.h    ${header_path}
     cp ggml/include/gguf.h         ${header_path}
+    cp tools/mtmd/mtmd.h           ${header_path}
+    cp tools/mtmd/mtmd-helper.h    ${header_path}
 
     # Create module map (common for all platforms)
     cat > ${module_path}module.modulemap << EOF
@@ -247,6 +251,7 @@ combine_static_libraries() {
         "${base_dir}/${build_dir}/ggml/src/${release_dir}/libggml-cpu.a"
         "${base_dir}/${build_dir}/ggml/src/ggml-metal/${release_dir}/libggml-metal.a"
         "${base_dir}/${build_dir}/ggml/src/ggml-blas/${release_dir}/libggml-blas.a"
+        "${base_dir}/${build_dir}/tools/mtmd/${release_dir}/libmtmd.a"
     )
 
     # Create temporary directory for processing
@@ -410,6 +415,7 @@ cmake -B build-ios-sim -G Xcode \
     -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
     -DLLAMA_OPENSSL=OFF \
+    -DMTMD_VIDEO=OFF \
     -S .
 cmake --build build-ios-sim --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
 
@@ -424,6 +430,7 @@ cmake -B build-ios-device -G Xcode \
     -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
     -DLLAMA_OPENSSL=OFF \
+    -DMTMD_VIDEO=OFF \
     -S .
 cmake --build build-ios-device --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
 
@@ -450,6 +457,7 @@ cmake -B build-visionos -G Xcode \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
     -DLLAMA_OPENSSL=OFF \
     -DLLAMA_BUILD_SERVER=OFF \
+    -DMTMD_VIDEO=OFF \
     -S .
 cmake --build build-visionos --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
 
@@ -465,6 +473,7 @@ cmake -B build-visionos-sim -G Xcode \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
     -DLLAMA_OPENSSL=OFF \
     -DLLAMA_BUILD_SERVER=OFF \
+    -DMTMD_VIDEO=OFF \
     -S .
 cmake --build build-visionos-sim --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
 
@@ -481,6 +490,7 @@ cmake -B build-tvos-sim -G Xcode \
     -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
     -DLLAMA_OPENSSL=OFF \
+    -DMTMD_VIDEO=OFF \
     -S .
 cmake --build build-tvos-sim --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
 
@@ -496,6 +506,7 @@ cmake -B build-tvos-device -G Xcode \
     -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
     -DLLAMA_OPENSSL=OFF \
+    -DMTMD_VIDEO=OFF \
     -S .
 cmake --build build-tvos-device --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet