server: (router) do not evict busy models

.github/workflows/build-apple.yml

@@ -60,7 +60,6 @@ jobs:
             -DCMAKE_BUILD_RPATH="@loader_path" \
             -DLLAMA_FATAL_WARNINGS=ON \
             -DLLAMA_BUILD_BORINGSSL=ON \
-            -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=OFF \
             -DGGML_METAL_SHADER_DEBUG=ON \
             -DGGML_RPC=ON \
@@ -127,7 +126,6 @@ jobs:
         run: |
           sysctl -a
           cmake -B build -G Xcode \
-            -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_OPENSSL=OFF \
             -DLLAMA_BUILD_APP=OFF \
@@ -178,7 +176,6 @@ jobs:
         run: |
           sysctl -a
           cmake -B build -G Xcode \
-            -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_BUILD_COMMON=OFF \
             -DLLAMA_BUILD_APP=OFF \
@@ -212,7 +209,6 @@ jobs:
         run: |
           sysctl -a
           cmake -B build -G Xcode \
-            -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_BUILD_COMMON=OFF \
             -DLLAMA_BUILD_APP=OFF \
@@ -257,7 +253,6 @@ jobs:
         run: |
           sysctl -a
           cmake -B build -G Xcode \
-            -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_OPENSSL=OFF \
             -DLLAMA_BUILD_APP=OFF \

.github/workflows/release.yml

@@ -93,13 +93,13 @@ jobs:
           - build: 'arm64'
             arch: 'arm64'
             os: macos-26
-            defines: "-DGGML_METAL_USE_BF16=ON -DGGML_METAL_EMBED_LIBRARY=ON -DCMAKE_OSX_DEPLOYMENT_TARGET=13.3"
+            defines: "-DGGML_METAL_EMBED_LIBRARY=ON -DCMAKE_OSX_DEPLOYMENT_TARGET=13.3"
           # TODO: this build is disabled to save Github Actions resources (https://github.com/ggml-org/llama.cpp/pull/23780)
           #       in order to enable it again, we have to provision dedicated runners  to run it
           #- build: 'arm64-kleidiai'
           #  arch: 'arm64'
           #  os: macos-14
-          #  defines: "-DGGML_METAL_USE_BF16=ON -DGGML_METAL_EMBED_LIBRARY=ON -DCMAKE_OSX_DEPLOYMENT_TARGET=13.3 -DGGML_CPU_KLEIDIAI=ON"
+          #  defines: "-DGGML_METAL_EMBED_LIBRARY=ON -DCMAKE_OSX_DEPLOYMENT_TARGET=13.3 -DGGML_CPU_KLEIDIAI=ON"
           - build: 'x64'
             arch: 'x64'
             os: macos-15-intel
@@ -1402,7 +1402,6 @@ jobs:
         run: |
           sysctl -a
           cmake -B build -G Xcode \
-            -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_OPENSSL=OFF \
             -DLLAMA_BUILD_APP=OFF \

build-xcframework.sh

@@ -17,7 +17,6 @@ LLAMA_BUILD_MTMD=ON
 GGML_METAL=ON
 GGML_METAL_EMBED_LIBRARY=ON
 GGML_BLAS_DEFAULT=ON
-GGML_METAL_USE_BF16=ON
 GGML_OPENMP=OFF
 
 COMMON_C_FLAGS="-Wno-macro-redefined -Wno-shorten-64-to-32 -Wno-unused-command-line-argument -g"
@@ -44,7 +43,6 @@ COMMON_CMAKE_ARGS=(
     -DGGML_METAL_EMBED_LIBRARY=${GGML_METAL_EMBED_LIBRARY}
     -DGGML_BLAS_DEFAULT=${GGML_BLAS_DEFAULT}
     -DGGML_METAL=${GGML_METAL}
-    -DGGML_METAL_USE_BF16=${GGML_METAL_USE_BF16}
     -DGGML_NATIVE=OFF
     -DGGML_OPENMP=${GGML_OPENMP}
 )

scripts/sync_vendor.py

@@ -24,10 +24,31 @@ vendor = {
     "https://raw.githubusercontent.com/sheredom/subprocess.h/8671cee1fc09f11a70ce3782a0ee13177c3aa387/subprocess.h": "vendor/sheredom/subprocess.h",
 }
 
+# TODO @ngxson : this is temporary, to be removed in the future
+patches = [
+    # https://github.com/sheredom/subprocess.h/pull/102
+    "vendor/sheredom/patch-bsd.patch",
+    # https://github.com/sheredom/subprocess.h/pull/101
+    "vendor/sheredom/patch-windows-quote-backslash.patch",
+    # https://github.com/sheredom/subprocess.h/pull/104
+    # note: must be applied after patch-bsd.patch, they touch adjacent lines
+    "vendor/sheredom/patch-glibc-older-than-2.29.patch",
+]
+
 for url, filename in vendor.items():
     print(f"downloading {url} to {filename}") # noqa: NP100
     urllib.request.urlretrieve(url, filename)
 
+for patch in patches:
+    print(f"applying {patch}") # noqa: NP100
+    try:
+        subprocess.check_call([
+            "git", "apply", "--directory", os.path.dirname(patch), patch
+        ])
+    except Exception as e:
+        print(f"Error: {e}") # noqa: NP100
+        sys.exit(1)
+
 print("Splitting httplib.h...") # noqa: NP100
 try:
     subprocess.check_call([

tools/server/server-models.cpp

@@ -721,7 +721,9 @@ void server_models::unload_lru() {
         for (const auto & m : mapping) {
             if (m.second.meta.is_running()) {
                 count_active++;
-                if (m.second.meta.last_used < lru_last_used) {
+                // do not evict busy one
+                bool is_model_idle = m.second.req_count == 0 && m.second.meta.is_ready_or_sleep();
+                if (is_model_idle && m.second.meta.last_used < lru_last_used) {
                     lru_model_name = m.first;
                     lru_last_used = m.second.meta.last_used;
                 }
@@ -739,6 +741,7 @@ void server_models::unload_lru() {
             });
         }
     }
+    // TODO @ngxson : if no idle model is found, queue the load request
 }
 
 void server_models::load(const std::string & name) {
@@ -1180,9 +1183,12 @@ server_http_res_ptr server_models::proxy_request(const server_http_req & req, co
     if (!meta->is_running()) {
         throw std::invalid_argument("model name=" + name + " is not running");
     }
-    if (update_last_used) {
+    {
         std::unique_lock<std::mutex> lk(mutex);
-        mapping[name].meta.last_used = ggml_time_ms();
+        if (update_last_used) {
+            mapping[name].meta.last_used = ggml_time_ms();
+        }
+        mapping[name].req_count++;
     }
     SRV_INF("proxying request to model %s on port %d\n", name.c_str(), meta->port);
     std::string proxy_path = req.path;
@@ -1198,13 +1204,22 @@ server_http_res_ptr server_models::proxy_request(const server_http_req & req, co
             req.headers,
             req.body,
             req.files,
-            // a detached request belongs to a replay session that outlives the client socket:
-            // it reaches the child even when the downstream died during the load wait, the
-            // session buffer is the recipient and DELETE remains the stop
-            detached ? std::function<bool()>([]() { return false; }) : req.should_stop,
+            // a detached request belongs to a replay session
+            detached
+                ? std::function<bool()>([]() { return false; })
+                : req.should_stop,
             base_params.timeout_read,
             base_params.timeout_write
             );
+
+    proxy->cleanup = [this, name]() {
+        std::unique_lock<std::mutex> lk(mutex);
+        auto it = mapping.find(name);
+        if (it != mapping.end() && it->second.req_count > 0) {
+            it->second.req_count--;
+        }
+    };
+
     return proxy;
 }
 
@@ -2064,7 +2079,7 @@ server_http_proxy::server_http_proxy(
     cli->set_write_timeout(timeout_read, 0); // reversed for cli (client) vs srv (server)
     cli->set_read_timeout(timeout_write, 0);
     this->status = 500; // to be overwritten upon response
-    this->cleanup = [pipe]() {
+    this->cleanup_pipes = [pipe]() {
         pipe->close_read();
         pipe->close_write();
     };

tools/server/server-models.h

@@ -84,7 +84,6 @@ struct server_model_meta {
     int exit_code = 0; // exit code of the model instance process (only valid if status == FAILED)
     int stop_timeout = 0; // seconds to wait before force-killing the model instance during shutdown
     mtmd_caps multimodal; // multimodal capabilities
-    // bool need_download = false; // whether the model needs to be downloaded before loading // TODO @ngxson: implement this
 
     bool is_ready() const {
         return status == SERVER_MODEL_STATUS_LOADED;
@@ -94,6 +93,10 @@ struct server_model_meta {
         return status == SERVER_MODEL_STATUS_LOADED || status == SERVER_MODEL_STATUS_LOADING || status == SERVER_MODEL_STATUS_SLEEPING;
     }
 
+    bool is_ready_or_sleep() const {
+        return status == SERVER_MODEL_STATUS_LOADED || status == SERVER_MODEL_STATUS_SLEEPING;
+    }
+
     bool is_failed() const {
         return status == SERVER_MODEL_STATUS_UNLOADED && exit_code != 0;
     }
@@ -113,6 +116,7 @@ private:
         std::shared_ptr<server_subproc> subproc; // shared between main thread and monitoring thread
         std::thread th;
         server_model_meta meta;
+        int req_count = 0; // number of active proxy requests
     };
 
     std::mutex mutex;
@@ -343,7 +347,6 @@ struct server_models_routes {
  */
 struct server_http_proxy : server_http_res {
     std::function<void()> cleanup = nullptr;
-public:
     server_http_proxy(const std::string & method,
                       const std::string & scheme,
                       const std::string & host,
@@ -357,11 +360,15 @@ public:
                       int32_t timeout_write
                       );
     ~server_http_proxy() {
+        if (cleanup_pipes) {
+            cleanup_pipes();
+        }
         if (cleanup) {
             cleanup();
         }
     }
 private:
+    std::function<void()> cleanup_pipes = nullptr;
     std::thread thread;
     struct msg_t {
         std::map<std::string, std::string> headers;

tools/ui/src/lib/constants/settings-registry.ts

@@ -198,7 +198,7 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
 				key: SETTINGS_KEYS.SHOW_MESSAGE_STATS,
 				label: 'Show message generation statistics',
 				help: 'Display generation statistics (tokens/second, token count, duration) below each assistant message.',
-				defaultValue: false,
+				defaultValue: true,
 				type: SettingsFieldType.CHECKBOX,
 				section: SETTINGS_SECTION_SLUGS.DISPLAY
 			},

vendor/sheredom/patch-bsd.patch

@@ -0,0 +1,19 @@
+Fix build on NetBSD, which provides posix_spawn_file_actions_addchdir()
+but not the _np() variant.
+
+Upstream PR: https://github.com/sheredom/subprocess.h/pull/102
+Applied locally by scripts/sync_vendor.py until it is merged upstream.
+
+diff --git a/subprocess.h b/subprocess.h
+index 5e809023a4..74a4e006c7 100644
+--- a/subprocess.h
++++ b/subprocess.h
+@@ -1205,7 +1205,7 @@ cleanup:
+ 
+   // Set working directory
+   if (process_cwd) {
+-#if defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED >= 260000
++#if defined(__NetBSD__) || (defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED >= 260000)
+     posix_error = posix_spawn_file_actions_addchdir(&actions, process_cwd);
+ #else
+ #if defined(__APPLE__) && defined(__clang__)

vendor/sheredom/patch-glibc-older-than-2.29.patch

@@ -0,0 +1,47 @@
+Fix building against glibc older than 2.29, which has no
+posix_spawn_file_actions_addchdir_np (the symbol is genuinely absent from
+libc.so, so no feature-test macro helps). Affects manylinux2014 (glibc 2.17)
+and manylinux_2_28, and was reported on RHEL 8.1. A requested process_cwd now
+fails with ENOSYS there instead of failing the build.
+
+Upstream PR: https://github.com/sheredom/subprocess.h/pull/104
+Applied locally by scripts/sync_vendor.py until it is merged upstream.
+
+(the README.md and test/ changes from the PR are omitted, we only vendor
+subprocess.h; rebased on top of patch-bsd.patch, so apply it after that one)
+
+diff --git a/subprocess.h b/subprocess.h
+index 1ef424a..c363393 100644
+--- a/subprocess.h
++++ b/subprocess.h
+@@ -274,6 +274,21 @@ subprocess_weak int subprocess_alive(struct subprocess_s *const process);
+ #include <unistd.h>
+ #endif
+ 
++/* Whether subprocess_create_ex can honour process_cwd. glibc only gained
++   posix_spawn_file_actions_addchdir_np in 2.29. Define this yourself to
++   override the detection, for instance on musl older than 1.1.24. */
++#if !defined(SUBPROCESS_HAVE_CWD)
++#if defined(__GLIBC__)
++#if __GLIBC_PREREQ(2, 29)
++#define SUBPROCESS_HAVE_CWD 1
++#else
++#define SUBPROCESS_HAVE_CWD 0
++#endif
++#else
++#define SUBPROCESS_HAVE_CWD 1
++#endif
++#endif
++
+ #if defined(_WIN32)
+ 
+ #include <wchar.h>
+@@ -1219,6 +1234,8 @@ cleanup:
+   if (process_cwd) {
+ #if defined(__NetBSD__) || (defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED >= 260000)
+     posix_error = posix_spawn_file_actions_addchdir(&actions, process_cwd);
++#elif !SUBPROCESS_HAVE_CWD
++    posix_error = ENOSYS;
+ #else
+ #if defined(__APPLE__) && defined(__clang__)
+ #pragma clang diagnostic push

vendor/sheredom/patch-windows-quote-backslash.patch

@@ -0,0 +1,107 @@
+Fix Windows command line quoting of backslash runs: a trailing backslash, or
+backslashes preceding a double quote, were not doubled, so CommandLineToArgvW
+in the child parsed them as escapes and mangled the argument list.
+
+Upstream PR: https://github.com/sheredom/subprocess.h/pull/101
+Applied locally by scripts/sync_vendor.py until it is merged upstream.
+
+(the test/ changes from the PR are omitted, we only vendor subprocess.h)
+
+diff --git a/subprocess.h b/subprocess.h
+index 5e80902..b06ad4d 100644
+--- a/subprocess.h
++++ b/subprocess.h
+@@ -653,6 +653,7 @@ int subprocess_create_ex(const char *const commandLine[], int options,
+   int wide_len;
+   int i, j;
+   int need_quoting;
++  subprocess_size_t bs_run;
+   unsigned long flags = 0;
+   unsigned long last_error = 0;
+   int result = subprocess_error_unknown;
+@@ -906,25 +907,29 @@ int subprocess_create_ex(const char *const commandLine[], int options,
+     len++;
+ 
+     // Quote the argument if it has a space in it
+-    if (strpbrk(commandLine[i], "\t\v ") != SUBPROCESS_NULL ||
+-        commandLine[i][0] == SUBPROCESS_NULL)
++    need_quoting = strpbrk(commandLine[i], "\t\v ") != SUBPROCESS_NULL ||
++                   commandLine[i][0] == SUBPROCESS_NULL;
++    if (need_quoting)
+       len += 2;
+ 
++    bs_run = 0;
+     for (j = 0; '\0' != commandLine[i][j]; j++) {
+-      switch (commandLine[i][j]) {
+-      default:
+-        break;
+-      case '\\':
+-        if (commandLine[i][j + 1] == '"') {
+-          len++;
+-        }
++      len++;
+ 
+-        break;
+-      case '"':
+-        len++;
+-        break;
++      if ('\\' == commandLine[i][j]) {
++        bs_run++;
++      } else {
++        if ('"' == commandLine[i][j]) {
++          // Duplicate the preceding run and escape the quote.
++          len += bs_run + 1;
++        }
++        bs_run = 0;
+       }
+-      len++;
++    }
++
++    if (need_quoting) {
++      // Duplicate trailing slashes before the generated closing quote.
++      len += bs_run;
+     }
+   }
+ 
+@@ -949,22 +954,29 @@ int subprocess_create_ex(const char *const commandLine[], int options,
+       commandLineCombined[len++] = '"';
+     }
+ 
+-    for (j = 0; '\0' != commandLine[i][j]; j++) {
+-      switch (commandLine[i][j]) {
+-      default:
+-        break;
+-      case '\\':
+-        if (commandLine[i][j + 1] == '"') {
+-          commandLineCombined[len++] = '\\';
+-        }
++    for (j = 0; '\0' != commandLine[i][j];) {
++      bs_run = 0;
++      while ('\\' == commandLine[i][j]) {
++        bs_run++;
++        j++;
++      }
++
++      if ('"' == commandLine[i][j]) {
++        // 2n + 1 slashes preserve n slashes and escape the quote.
++        bs_run = (bs_run * 2) + 1;
++      } else if ('\0' == commandLine[i][j] && need_quoting) {
++        // 2n slashes preserve n slashes before the closing quote.
++        bs_run *= 2;
++      }
+ 
+-        break;
+-      case '"':
++      while (bs_run > 0) {
+         commandLineCombined[len++] = '\\';
+-        break;
++        bs_run--;
+       }
+ 
+-      commandLineCombined[len++] = commandLine[i][j];
++      if ('\0' != commandLine[i][j]) {
++        commandLineCombined[len++] = commandLine[i][j++];
++      }
+     }
+     if (need_quoting) {
+       commandLineCombined[len++] = '"';

vendor/sheredom/subprocess.h

@@ -274,6 +274,21 @@ subprocess_weak int subprocess_alive(struct subprocess_s *const process);
 #include <unistd.h>
 #endif
 
+/* Whether subprocess_create_ex can honour process_cwd. glibc only gained
+   posix_spawn_file_actions_addchdir_np in 2.29. Define this yourself to
+   override the detection, for instance on musl older than 1.1.24. */
+#if !defined(SUBPROCESS_HAVE_CWD)
+#if defined(__GLIBC__)
+#if __GLIBC_PREREQ(2, 29)
+#define SUBPROCESS_HAVE_CWD 1
+#else
+#define SUBPROCESS_HAVE_CWD 0
+#endif
+#else
+#define SUBPROCESS_HAVE_CWD 1
+#endif
+#endif
+
 #if defined(_WIN32)
 
 #include <wchar.h>
@@ -653,6 +668,7 @@ int subprocess_create_ex(const char *const commandLine[], int options,
   int wide_len;
   int i, j;
   int need_quoting;
+  subprocess_size_t bs_run;
   unsigned long flags = 0;
   unsigned long last_error = 0;
   int result = subprocess_error_unknown;
@@ -906,25 +922,29 @@ int subprocess_create_ex(const char *const commandLine[], int options,
     len++;
 
     // Quote the argument if it has a space in it
-    if (strpbrk(commandLine[i], "\t\v ") != SUBPROCESS_NULL ||
-        commandLine[i][0] == SUBPROCESS_NULL)
+    need_quoting = strpbrk(commandLine[i], "\t\v ") != SUBPROCESS_NULL ||
+                   commandLine[i][0] == SUBPROCESS_NULL;
+    if (need_quoting)
       len += 2;
 
+    bs_run = 0;
     for (j = 0; '\0' != commandLine[i][j]; j++) {
-      switch (commandLine[i][j]) {
-      default:
-        break;
-      case '\\':
-        if (commandLine[i][j + 1] == '"') {
-          len++;
-        }
-
-        break;
-      case '"':
-        len++;
-        break;
-      }
       len++;
+
+      if ('\\' == commandLine[i][j]) {
+        bs_run++;
+      } else {
+        if ('"' == commandLine[i][j]) {
+          // Duplicate the preceding run and escape the quote.
+          len += bs_run + 1;
+        }
+        bs_run = 0;
+      }
+    }
+
+    if (need_quoting) {
+      // Duplicate trailing slashes before the generated closing quote.
+      len += bs_run;
     }
   }
 
@@ -949,22 +969,29 @@ int subprocess_create_ex(const char *const commandLine[], int options,
       commandLineCombined[len++] = '"';
     }
 
-    for (j = 0; '\0' != commandLine[i][j]; j++) {
-      switch (commandLine[i][j]) {
-      default:
-        break;
-      case '\\':
-        if (commandLine[i][j + 1] == '"') {
-          commandLineCombined[len++] = '\\';
-        }
-
-        break;
-      case '"':
-        commandLineCombined[len++] = '\\';
-        break;
+    for (j = 0; '\0' != commandLine[i][j];) {
+      bs_run = 0;
+      while ('\\' == commandLine[i][j]) {
+        bs_run++;
+        j++;
       }
 
-      commandLineCombined[len++] = commandLine[i][j];
+      if ('"' == commandLine[i][j]) {
+        // 2n + 1 slashes preserve n slashes and escape the quote.
+        bs_run = (bs_run * 2) + 1;
+      } else if ('\0' == commandLine[i][j] && need_quoting) {
+        // 2n slashes preserve n slashes before the closing quote.
+        bs_run *= 2;
+      }
+
+      while (bs_run > 0) {
+        commandLineCombined[len++] = '\\';
+        bs_run--;
+      }
+
+      if ('\0' != commandLine[i][j]) {
+        commandLineCombined[len++] = commandLine[i][j++];
+      }
     }
     if (need_quoting) {
       commandLineCombined[len++] = '"';
@@ -1205,8 +1232,10 @@ cleanup:
 
   // Set working directory
   if (process_cwd) {
-#if defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED >= 260000
+#if defined(__NetBSD__) || (defined(__APPLE__) && MAC_OS_X_VERSION_MIN_REQUIRED >= 260000)
     posix_error = posix_spawn_file_actions_addchdir(&actions, process_cwd);
+#elif !SUBPROCESS_HAVE_CWD
+    posix_error = ENOSYS;
 #else
 #if defined(__APPLE__) && defined(__clang__)
 #pragma clang diagnostic push