nits

tools/server/README.md

@@ -1757,7 +1757,7 @@ The precedence rule for preset options is as follows:
 3. **Global options** defined in the preset file (`[*]`)
 
 We also offer additional options that are exclusive to presets (these aren't treated as command-line arguments):
-- `load-on-startup` (boolean): Controls whether the model loads automatically when the server starts
+- `load-on-startup` (boolean): Controls whether the model loads automatically when the server starts. Only applies at startup: if the model list is reloaded later (for example after editing the preset file), a newly added model is listed but not loaded
 - `stop-timeout` (int, seconds): After requested unload, wait for this many seconds before forcing termination (default: 10)
 - `dedup-cache-models` (boolean): When the preset uses `hf-repo` pointing to a model that is already downloaded, hide the corresponding cached model entry from `GET /models` (the preset entry remains visible). Set it in the `[*]` section to apply to all presets.
 

tools/server/server-models.cpp

@@ -822,7 +822,6 @@ void server_models::load_models() {
         }
 
         // add models that are new in this reload
-        std::vector<std::string> newly_added;
         for (const auto & [name, preset] : final_presets) {
             if (mapping.find(name) == mapping.end()) {
                 server_model_meta meta{
@@ -843,37 +842,20 @@ void server_models::load_models() {
                     // /* need_download */ false,
                 };
                 add_model(std::move(meta));
-                newly_added.push_back(name);
             }
         }
 
         apply_stop_timeout();
         apply_hidden();
 
-        // clear reload flag before unlocking for autoload - load() blocks on !is_reloading,
-        // so clearing it here (while still locked) prevents a deadlock in the autoload calls below
+        // clear reload flag before unlocking - load() blocks on !is_reloading
         is_reloading = false;
         cv.notify_all();
 
         log_available_models();
 
-        // collect autoload candidates while still under the lock
-        std::vector<std::string> to_autoload;
-        for (const auto & name : newly_added) {
-            auto it = mapping.find(name);
-            if (it != mapping.end()) {
-                std::string val;
-                if (it->second.meta.preset.get_option(COMMON_ARG_PRESET_LOAD_ON_STARTUP, val) && common_arg_utils::is_truthy(val)) {
-                    to_autoload.push_back(name);
-                }
-            }
-        }
-
+        // note: load-on-startup is not honored here, a reload never spawns an instance
         lk.unlock();
-        for (const auto & name : to_autoload) {
-            SRV_INF("(reload) loading new model %s\n", name.c_str());
-            load(name);
-        }
 
         notify_sse("models_reload", "*");
     }

tools/server/server-models.h

@@ -137,7 +137,7 @@ private:
     bool need_reload = false;
 
     // models marked with load-on-startup, unset once load_startup_models() drains it
-    // using std::optional to make sure it's only called once after is_first_load, load_startup_models will unset it
+    // no value means the startup phase is over, so a reload must not queue anything
     std::optional<std::vector<std::string>> startup_models{std::in_place};
 
     // conv_id -> model name that currently serves its stream session, lets the resumable stream