Merge commit '9d5d882d8cd0f0a9283d87ed5e6fe3ee0d925fb1' into concedo_experimental

# Conflicts:
#	.github/labeler.yml
#	app/CMakeLists.txt
#	app/llama.cpp
#	build-xcframework.sh
#	common/CMakeLists.txt
#	common/download.h
#	docs/backend/SYCL.md
#	docs/backend/snapdragon/CMakeUserPresets.json
#	docs/speculative.md
#	ggml/CMakeLists.txt
#	ggml/include/ggml-sycl.h
#	ggml/src/ggml-hexagon/CMakeLists.txt
#	ggml/src/ggml-hexagon/ggml-hexagon.cpp
#	ggml/src/ggml-hexagon/htp/CMakeLists.txt
#	ggml/src/ggml-hexagon/htp/cmake-toolchain.cmake
#	ggml/src/ggml-hexagon/htp/flash-attn-ops.c
#	ggml/src/ggml-hexagon/htp/hex-dma.h
#	ggml/src/ggml-hexagon/htp/hex-utils.h
#	ggml/src/ggml-hexagon/htp/hmx-flash-attn-ops.c
#	ggml/src/ggml-hexagon/htp/htp-ctx.h
#	ggml/src/ggml-hexagon/htp/htp-ops.h
#	ggml/src/ggml-hexagon/htp/htp_iface.idl
#	ggml/src/ggml-hexagon/htp/hvx-base.h
#	ggml/src/ggml-hexagon/htp/main.c
#	ggml/src/ggml-hexagon/htp/matmul-ops.c
#	ggml/src/ggml-hexagon/libggml-htp.inf
#	ggml/src/ggml-opencl/ggml-opencl.cpp
#	ggml/src/ggml-opencl/kernels/norm.cl
#	ggml/src/ggml-sycl/conv3d.cpp
#	ggml/src/ggml-sycl/ggml-sycl.cpp
#	scripts/snapdragon/adb/run-completion.sh
#	scripts/snapdragon/adb/run-tool.sh
#	scripts/snapdragon/ggml-hexagon-profile.py
#	tests/CMakeLists.txt
#	tests/test-backend-ops.cpp
#	tests/test-thread-safety.cpp
#	tools/llama-bench/llama-bench.cpp
#	tools/mtmd/CMakeLists.txt
#	tools/mtmd/tests/test-deepseek-ocr.py

tools/server/server-models.cpp

@@ -223,8 +223,8 @@ void server_model_meta::update_caps() {
             "LLAMA_ARG_HF_REPO_FILE",
         });
         params.offline = true;
-        // params.skip_download = true; // TODO: ideally, we should validate the model here, but it takes too much time
-        common_params_handle_models(params, LLAMA_EXAMPLE_SERVER, {});
+        common_models_handler handler = common_models_handler_init(params, LLAMA_EXAMPLE_SERVER);
+        common_models_handler_apply(handler, params); // note: this won't download the model because offline=true
         if (params.mmproj.path.empty()) {
             multimodal = { false, false };
         } else {
@@ -1393,9 +1393,8 @@ struct server_download_state : public common_download_callback {
 
     bool run(common_params & params) {
         try {
-            common_params_handle_models_params p;
-            p.callback = this;
-            common_params_handle_models(params, LLAMA_EXAMPLE_SERVER, p);
+            common_models_handler handler = common_models_handler_init(params, LLAMA_EXAMPLE_SERVER);
+            common_models_handler_apply(handler, params, this);
             is_ok = true;
         } catch (const std::exception & e) {
             auto model_name = params.model.get_name();
@@ -1768,23 +1767,14 @@ void server_models_routes::init_routes() {
             throw std::invalid_argument("model must be a non-empty string");
         }
 
-        common_params_model model;
-        common_download_opts opts;
+        common_params p;
+        p.model.hf_repo  = name;
+        p.hf_token       = params.hf_token;
 
-        model.hf_repo        = name;
-        opts.bearer_token    = params.hf_token;
-        // note: we only check main model, no need sidecar here
-        opts.download_mmproj = false;
-        opts.download_mtp    = false;
-
-        // first, only check if the model is valid and can be downloaded
-        opts.skip_download = true;
+        // validate by fetching metadata
         bool ok = false;
         try {
-            auto validation = common_download_model(model, opts);
-            ok = !validation.model_path.empty();
-        } catch (const common_skip_download_exception &) {
-            // model is valid and will be downloaded
+            common_models_handler_init(p, LLAMA_EXAMPLE_SERVER);
             ok = true;
         } catch (...) {
             SRV_ERR("unknown error while validating model '%s'\n", name.c_str());

tools/server/server.cpp

@@ -89,15 +89,16 @@ int llama_server(int argc, char ** argv) {
     llama_backend_init();
     llama_numa_init(params.numa);
 
-    // note: router mode also accepts -hf remote-preset, so we need to check that first
-    if (!params.model.hf_repo.empty()) {
-        try {
-            common_params_handle_models_params handle_params;
-            handle_params.preset_only = true;
-            common_params_handle_models(params, LLAMA_EXAMPLE_SERVER, handle_params);
-        } catch (const std::exception & e) {
-            // ignored for now
+    common_models_handler models_handler;
+    try {
+        models_handler = common_models_handler_init(params, LLAMA_EXAMPLE_SERVER);
+        if (common_models_handler_is_preset_repo(models_handler)) {
+            // apply the preset and start the server in router mode
+            common_models_handler_apply(models_handler, params);
         }
+    } catch (const std::exception & e) {
+        SRV_ERR("failed to fetch model metadata: %s\n", e.what());
+        return 1;
     }
 
     // router server never loads a model and must not touch the GPU
@@ -241,6 +242,19 @@ int llama_server(int argc, char ** argv) {
     // Google Cloud Platform (Vertex AI) compat
     ctx_http.register_gcp_compat();
 
+    // return 403 for disabled features
+    server_http_context::handler_t res_403 = [](const server_http_req &) {
+        auto res = std::make_unique<server_http_res>();
+        res->status = 403;
+        res->data = safe_json_to_str({
+            {"error", {
+                {"message", "this feature is disabled"},
+                {"type", "feature_disabled"},
+            }}
+        });
+        return res;
+    };
+
     // CORS proxy (EXPERIMENTAL, only used by the Web UI for MCP)
     if (params.ui_mcp_proxy) {
         SRV_WRN("%s", "-----------------\n");
@@ -249,7 +263,11 @@ int llama_server(int argc, char ** argv) {
         SRV_WRN("%s", "-----------------\n");
         ctx_http.get ("/cors-proxy",      ex_wrapper(proxy_handler_get));
         ctx_http.post("/cors-proxy",      ex_wrapper(proxy_handler_post));
+    } else {
+        ctx_http.get ("/cors-proxy",      ex_wrapper(res_403));
+        ctx_http.post("/cors-proxy",      ex_wrapper(res_403));
     }
+
     // EXPERIMENTAL built-in tools
     if (!params.server_tools.empty()) {
         try {
@@ -264,6 +282,9 @@ int llama_server(int argc, char ** argv) {
         SRV_WRN("%s", "-----------------\n");
         ctx_http.get ("/tools",           ex_wrapper(tools.handle_get));
         ctx_http.post("/tools",           ex_wrapper(tools.handle_post));
+    } else {
+        ctx_http.get ("/tools",           ex_wrapper(res_403));
+        ctx_http.post("/tools",           ex_wrapper(res_403));
     }
 
     //
@@ -274,7 +295,12 @@ int llama_server(int argc, char ** argv) {
         return child.run_download(params);
     } else if (!is_router_server) {
         // single-model mode (NOT spawned by router)
-        common_params_handle_models(params, LLAMA_EXAMPLE_SERVER, {});
+        try {
+            common_models_handler_apply(models_handler, params);
+        } catch (const std::exception & e) {
+            SRV_ERR("failed to download model: %s\n", e.what());
+            return 1;
+        }
     }
 
     //

tools/server/tests/unit/test_proxy.py

@@ -16,7 +16,7 @@ def test_mcp_no_proxy():
     server.start()
 
     res = server.make_request("GET", "/cors-proxy")
-    assert res.status_code == 404
+    assert res.status_code == 403
 
 
 def test_mcp_proxy():