feat: bump OpenMoss engine to 0.2.0, fix cascade generation, add OpenMOSS SFX model

docs/dev/backends-reference.md

@@ -283,12 +283,13 @@ the generator instead. Prose outside the markers is preserved. -->
 | `Bert-Phishing-ONNX` | 1.34 | classification |
 | `Phishing-Email-Detection-ONNX` | 0.27 | classification |
 
-#### `openmoss` — OpenMOSS TTS (2 models)
+#### `openmoss` — OpenMOSS TTS (3 models)
 
 | Model | Size (GB) | Labels |
 |-------|-----------|--------|
-| `MOSS-VoiceGen` | 7.3 | tts, voice-design |
-| `OpenMOSS-TTS` | 12.5 | tts |
+| `MOSS-SoundEffect` | 6.6 | audio-generation |
+| `MOSS-TTS-Local` | 15.8 | tts |
+| `OpenMOSS-TTS` | 19.8 | tts |
 
 #### `ryzenai-llm` — Ryzen AI LLM (79 models)
 

src/app/src/App.tsx

@@ -6,6 +6,7 @@ import { AccountSession, currentSession, subscribeAccountSessionChanges } from '
 import { setPresetStorageScope } from './presetStore';
 import { customModelToModelInfo, loadCustomModels } from './features/customModels/customModelStore';
 import { findModelInfoByName, isCollectionFullyLoaded, isCollectionModel, withVirtualLoadedCollections } from './features/collections/collectionModels';
+import { isModelSelectionLocked, onModelSelectionUnlocked } from './features/modelSettings/modelSelectionLock';
 import ChatView from './components/ChatView';
 import ModelManager from './components/ModelManager';
 import ConnectView from './components/ConnectView';
@@ -314,10 +315,16 @@ const App: React.FC = () => {
     };
     setLoadedModels(enriched);
     setCurrentModel(current => {
+      if (current && isModelSelectionLocked()) return current;
       if (current && enriched.some(m => m.model_name === current && (canSelectInComposer(m) || customSelectable(m.model_name) || infoSelectable(m.model_name)))) return current;
       if (current) {
         const info = findModelInfoByName(knownInfos, current);
         if (info && isCollectionModel(info) && isCollectionFullyLoaded(info, loaded)) return current;
+        // A downloaded model the user picked stays picked while it is still
+        // being prepared. Loading a backend can take minutes (install, then
+        // weights), and eviction is a resource decision, not a user one —
+        // neither should silently retarget the composer.
+        if (info && (info as any).downloaded) return current;
       }
       const virtualOmni = enriched.find(model => {
         const info = findModelInfoByName(knownInfos, model.model_name);
@@ -330,6 +337,12 @@ const App: React.FC = () => {
     });
   }, [accountSession.storageScope]);
 
+  useEffect(() => onModelSelectionUnlocked(() => {
+    api.refresh()
+      .then(result => { if (result) applyLoadedModels(result.health.all_models_loaded); })
+      .catch(() => { });
+  }), [applyLoadedModels]);
+
   const navigateToRoute = useCallback((nextRoute: AppRoute) => {
     if (nextRoute.view === 'dashboard') lastWorkspaceSectionsRef.current.dashboard = nextRoute.section;
     if (nextRoute.view === 'connect') lastWorkspaceSectionsRef.current.connect = nextRoute.section;

src/app/src/features/modelSettings/modelSelectionLock.ts

@@ -0,0 +1,27 @@
+type Listener = () => void;
+
+let depth = 0;
+const listeners = new Set<Listener>();
+
+export function isModelSelectionLocked(): boolean {
+  return depth > 0;
+}
+
+export function onModelSelectionUnlocked(listener: Listener): () => void {
+  listeners.add(listener);
+  return () => { listeners.delete(listener); };
+}
+
+export async function withModelSelectionLock<T>(run: () => Promise<T>): Promise<T> {
+  depth += 1;
+  try {
+    return await run();
+  } finally {
+    depth = Math.max(0, depth - 1);
+    if (depth === 0) {
+      listeners.forEach(listener => {
+        try { listener(); } catch { }
+      });
+    }
+  }
+}

src/app/src/styles/styles.css

@@ -11215,6 +11215,34 @@ button.detail-presets__explanation-step {
   cursor: not-allowed;
 }
 
+.composer__openmoss-params {
+  min-width: 0;
+  flex-basis: 100%;
+}
+
+.composer__openmoss-params > summary {
+  cursor: pointer;
+  font-size: var(--type-overline-size);
+  line-height: 1;
+  letter-spacing: 0.08em;
+  text-transform: uppercase;
+  font-weight: var(--weight-semibold);
+  color: var(--text-tertiary);
+  padding: var(--space-1) 0;
+  user-select: none;
+}
+
+.composer__openmoss-params > summary:hover {
+  color: var(--text-secondary);
+}
+
+.composer__openmoss-params-grid {
+  display: grid;
+  grid-template-columns: repeat(auto-fit, minmax(96px, 1fr));
+  gap: var(--space-2) var(--space-3);
+  padding-top: var(--space-2);
+}
+
 .composer__openmoss-status {
   grid-column: 1 / -1;
   min-height: 18px;

src/cpp/include/lemon/backends/openmoss/openmoss_server.h

@@ -5,12 +5,18 @@
 #include "lemon/wrapped_server.h"
 #include "lemon/server_capabilities.h"
 #include "lemon/backends/backend_utils.h"
+#include <map>
+#include <mutex>
 #include <string>
+#include <utility>
+#include <vector>
 
 namespace lemon {
 namespace backends {
 
-class OpenMossServer : public WrappedServer, public ITextToSpeechServer {
+class OpenMossServer : public WrappedServer,
+                       public ITextToSpeechServer,
+                       public IAudioGenerationServer {
 public:
     static InstallParams get_install_params(const std::string& backend, const std::string& version);
 
@@ -26,10 +32,30 @@ class OpenMossServer : public WrappedServer, public ITextToSpeechServer {
     void unload() override;
 
     void audio_speech(const json& request, httplib::DataSink& sink) override;
+    void audio_generations(const json& request, httplib::DataSink& sink) override;
     std::vector<std::string> supported_audio_formats() const override { return {"wav"}; }
 
 private:
+    struct Subprocess {
+        ProcessHandle handle;
+        int port = 0;
+    };
+
     std::string resolve_binary_path(const std::string& backend);
+    Subprocess spawn(const std::string& model_path);
+
+    // The voice-design model is a component of the speech model, but
+    // moss-tts-server hosts exactly one --model per process. It is spawned only
+    // to render a reference sample and torn down immediately, so steady-state
+    // residency stays at the speech model alone.
+    std::string design_reference_sample(const std::string& voice_description);
+    json apply_voice_design(const json& request);
+
+    std::string exe_path_;
+    std::string voicegen_path_;
+    std::map<std::string, std::string> reference_cache_;
+    std::vector<std::pair<std::string, std::string>> env_vars_;
+    std::mutex design_mutex_;
 };
 
 namespace openmoss {

src/cpp/resources/backend_versions.json

@@ -11,6 +11,14 @@
           "moss-tts-rocm-windows-x64.zip": "sha256:c9178dfad4ddf51fc8d8df8d1412e65237ff364ca9ca8fc997c9c71f55d60827",
           "moss-tts-cuda-linux-x64.tar.gz": "sha256:8662251a854bda4993a5f99e90ba02c176797bf22b1c5b3e89a0061b0a4b102a",
           "moss-tts-cuda-windows-x64.zip": "sha256:2b2f5f4b6757d99acf621b43426483ab0bc4a70f1f13904c17deed09cac70969"
+        },
+        "v0.2.0": {
+          "moss-tts-vulkan-linux-x64.tar.gz": "sha256:26f67337940cffaa26efd233764888c1e571ff438b16a8c432ca090e31e3d372",
+          "moss-tts-vulkan-windows-x64.zip": "sha256:d35ff25c39b0f9fb75191a8d267e4350fdb9bad8ed68f23f2925272a4d27354c",
+          "moss-tts-rocm-linux-x64.tar.gz": "sha256:32b94cc4b3c7fd6856ed6f73f281ac2099eacc07dfd3344ae34ee4145e0a88d0",
+          "moss-tts-rocm-windows-x64.zip": "sha256:17e782acafed4865a40ed2f8330c46dbdec15cff5d6cb66c243feeaa5d3e5ad0",
+          "moss-tts-cuda-linux-x64.tar.gz": "sha256:f4b7ab3fc308c66c984ee42fa0b038b483dde2d20dbfed11fa7f94d78205365c",
+          "moss-tts-cuda-windows-x64.zip": "sha256:8229a74e51f111db3214b3e3c3fd2d9e6dad34443448693023519c4bb7da688d"
         }
       },
       "pwilkin/acestep.cpp": {
@@ -182,9 +190,9 @@
     "cuda": "v0.4.3"
   },
   "openmoss": {
-    "vulkan": "v0.1.3",
-    "rocm-stable": "v0.1.3",
-    "cuda": "v0.1.3"
+    "vulkan": "v0.2.0",
+    "rocm-stable": "v0.2.0",
+    "cuda": "v0.2.0"
   },
   "clear_bin_if_lemonade_below": "9.4.0"
 }

src/cpp/resources/server_models.json

@@ -1657,23 +1657,70 @@
         "size": 15.4
     },
     "OpenMOSS-TTS": {
-        "checkpoint": "ilintar/moss-tts-gguf:moss-tts-1.5-q8_0.gguf",
+        "checkpoints": {
+            "main": "ilintar/moss-tts-gguf:moss-tts-1.5-q8_0.gguf",
+            "main_extras": "ilintar/moss-tts-gguf:moss-tts-1.5-q8_0.extras.gguf",
+            "voicegen": "ilintar/moss-voicegen-gguf:moss-voicegen.gguf",
+            "voicegen_extras": "ilintar/moss-voicegen-gguf:moss-voicegen.extras.gguf"
+        },
+        "recipe": "openmoss",
+        "suggested": true,
+        "labels": [
+            "tts"
+        ],
+        "size": 19.8,
+        "speech_defaults": {
+            "audio_temperature": 1.7,
+            "audio_top_p": 0.8,
+            "audio_top_k": 25,
+            "audio_repetition_penalty": 1.0,
+            "text_temperature": 1.5,
+            "text_top_p": 1.0,
+            "text_top_k": 50,
+            "speed": 1.0
+        }
+    },
+    "MOSS-TTS-Local": {
+        "checkpoints": {
+            "main": "ilintar/moss-tts-local-gguf:moss-tts-local-1.5-q8_0.gguf",
+            "main_extras": "ilintar/moss-tts-local-gguf:moss-tts-local-1.5-q8_0.extras.gguf",
+            "voicegen": "ilintar/moss-voicegen-gguf:moss-voicegen.gguf",
+            "voicegen_extras": "ilintar/moss-voicegen-gguf:moss-voicegen.extras.gguf"
+        },
         "recipe": "openmoss",
         "suggested": true,
         "labels": [
             "tts"
         ],
-        "size": 12.5
+        "size": 15.8,
+        "speech_defaults": {
+            "audio_temperature": 1.5,
+            "audio_top_p": 0.6,
+            "audio_top_k": 50,
+            "audio_repetition_penalty": 1.1,
+            "text_temperature": 1.5,
+            "text_top_p": 1.0,
+            "text_top_k": 50,
+            "speed": 1.0
+        }
     },
-    "MOSS-VoiceGen": {
-        "checkpoint": "ilintar/moss-voicegen-gguf:moss-voicegen.gguf",
+    "MOSS-SoundEffect": {
+        "checkpoints": {
+            "main": "ilintar/moss-soundeffect-gguf:moss-soundeffect-2.0.gguf",
+            "main_extras": "ilintar/moss-soundeffect-gguf:moss-soundeffect-2.0.extras.gguf"
+        },
         "recipe": "openmoss",
         "suggested": true,
         "labels": [
-            "tts",
-            "voice-design"
+            "audio-generation"
         ],
-        "size": 7.3
+        "size": 6.6,
+        "audio_defaults": {
+            "seconds": 10,
+            "steps": 100,
+            "cfg_scale": 4.0,
+            "sigma_shift": 5.0
+        }
     },
     "RealESRGAN-x4plus": {
         "checkpoint": "amd/realesrgan-x4plus:RealESRGAN_x4plus.pth",