Merge branch 'upstream' into concedo_experimental

# Conflicts:
#	.github/workflows/build-apple.yml
#	.github/workflows/build-cmake-pkg.yml
#	.github/workflows/release.yml
#	.pi/gg/SYSTEM.md
#	CMakeLists.txt
#	CODEOWNERS
#	README.md
#	build-xcframework.sh
#	ci/run.sh
#	docs/build.md
#	examples/CMakeLists.txt
#	examples/llama.android/lib/build.gradle.kts
#	ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_tile.wgsl
#	tests/CMakeLists.txt
#	tests/test-backend-ops.cpp
#	tests/test-save-load-state.cpp
#	tools/batched-bench/CMakeLists.txt
#	tools/cli/CMakeLists.txt
#	tools/completion/CMakeLists.txt
#	tools/llama-bench/CMakeLists.txt
#	tools/perplexity/CMakeLists.txt
#	tools/quantize/CMakeLists.txt
#	tools/server/CMakeLists.txt

tools/fit-params/CMakeLists.txt

@@ -2,11 +2,16 @@
 
 set(TARGET llama-fit-params-impl)
 
-add_library(${TARGET} STATIC fit-params.cpp)
+add_library(${TARGET} fit-params.cpp)
+set_target_properties(${TARGET} PROPERTIES WINDOWS_EXPORT_ALL_SYMBOLS ON)
 
 target_include_directories(${TARGET} PUBLIC ${CMAKE_CURRENT_SOURCE_DIR})
 target_link_libraries(${TARGET} PUBLIC llama-common llama ${CMAKE_THREAD_LIBS_INIT})
 
+if(LLAMA_TOOLS_INSTALL)
+    install(TARGETS ${TARGET} LIBRARY)
+endif()
+
 # llama-fit-params executable
 
 set(TARGET llama-fit-params)

tools/server/server-context.cpp

@@ -506,6 +506,9 @@ struct server_slot {
 
         if (ptask) {
             res["id_task"] = ptask->id;
+            res["n_prompt_tokens"]           = (int32_t) prompt.tokens.size();
+            res["n_prompt_tokens_processed"] = n_prompt_tokens_processed;
+            res["n_prompt_tokens_cache"]     = n_prompt_tokens_cache;
             res["params"] = ptask->params.to_json(only_metrics);
             res["next_token"] = {
                 {
@@ -701,6 +704,10 @@ private:
     bool sleeping = false;
 
     void destroy() {
+        spec.reset();
+        ctx_dft.reset();
+        model_dft.reset();
+
         llama_init.reset();
 
         ctx_tgt = nullptr;

tools/server/server-models.cpp

@@ -14,6 +14,7 @@
 #include <mutex>
 #include <condition_variable>
 #include <cstring>
+#include <cstdlib>
 #include <atomic>
 #include <chrono>
 #include <queue>
@@ -159,6 +160,13 @@ void server_model_meta::update_args(common_preset_context & ctx_preset, std::str
     // TODO: maybe validate preset before rendering ?
     // render args
     args = preset.to_args(bin_path);
+
+    // unified binary dispatches by subcommand, re-inject it right after the
+    // binary path so the child starts as 'llama serve ...' not 'llama ...'
+    const char * app_cmd = std::getenv("LLAMA_APP_CMD");
+    if (app_cmd != nullptr && app_cmd[0] != '\0' && !bin_path.empty()) {
+        args.insert(args.begin() + 1, app_cmd);
+    }
 }
 
 void server_model_meta::update_caps() {