Check Point

daemon/qvac-master.js

@@ -1,3 +1,5 @@
+import { voiceSettings } from './voice-settings.js';
+import { profile } from './model-profiles.js';
 import { createRequire } from 'node:module';
 import fs from 'node:fs';
 import path from 'node:path';
@@ -17,7 +19,7 @@ const auxiliaryModels = new Map();
 
 export const QVAC_MASTER = Object.freeze({
   configPath: process.env.QVAC_CONFIG_PATH,
-  model: process.env.JARVIS_QVAC_MODEL || 'qwen3.5-4b',
+  model: process.env.JARVIS_QVAC_MODEL || voiceSettings().chatModel || profile(voiceSettings().modelProfile).model,
   device: 'gpu',
   gpuLayers: 99,
 });
@@ -45,34 +47,26 @@ export function assertSdkVersion() {
 
 export async function acquireQvac({ auxiliaryOnly = false } = {}) {
   if (auxiliaryOnly) { await qvacSdk(); ownerCount += 1; return; }
-  ownerCount += 1;
   if (!loadPromise) {
-    assertSdkVersion();
-    const resources = await Agent.engine.resources();
-    const gpuVisible = (resources.gpus?.length || 0) > 0 ||
-      Boolean(resources.drivers?.vulkan || resources.drivers?.cuda || resources.drivers?.opencl || resources.gpu);
-    if (!gpuVisible) {
-      ownerCount = Math.max(0, ownerCount - 1);
-      throw new Error('Jarvis requires a QVAC-visible GPU backend; run npm run gpu-doctor');
-    }
-    loadPromise = Agent.engine.load({
-      model: QVAC_MASTER.model,
-      tools: true,
-      device: 'gpu',
-      gpu_layers: QVAC_MASTER.gpuLayers,
-      mmprojUseGpu: true,
-    }).then((loaded) => {
-      if (loaded.device !== 'gpu') {
-        throw new Error(`Jarvis requires GPU QVAC inference; loaded device was ${loaded.device}`);
-      }
+    // Publish the promise before the first await so concurrent callers share
+    // resource discovery and loading. Count only successful acquisitions.
+    loadPromise = Promise.resolve().then(async () => {
+      assertSdkVersion();
+      const resources = await Agent.engine.resources();
+      const gpuVisible = (resources.gpus?.length || 0) > 0 ||
+        Boolean(resources.drivers?.vulkan || resources.drivers?.cuda || resources.drivers?.opencl || resources.gpu);
+      if (!gpuVisible) throw new Error('Jarvis requires a QVAC-visible GPU backend; run npm run gpu-doctor');
+      const loaded = await Agent.engine.load({
+        model: QVAC_MASTER.model, tools: true, device: 'gpu',
+        gpu_layers: QVAC_MASTER.gpuLayers, mmprojUseGpu: true,
+      });
+      if (loaded.device !== 'gpu') throw new Error(`Jarvis requires GPU QVAC inference; loaded device was ${loaded.device}`);
       return loaded;
-    }).catch((error) => {
-      loadPromise = null;
-      ownerCount = Math.max(0, ownerCount - 1);
-      throw error;
-    });
+    }).catch((error) => { loadPromise = null; throw error; });
   }
-  return loadPromise;
+  const loaded = await loadPromise;
+  ownerCount += 1;
+  return loaded;
 }
 
 export async function qvacSdk() {
@@ -92,7 +86,7 @@ function resolveSdkAsset(sdk, name) {
 
 function resolveModelConfigAssets(sdk, config) {
   const copy = { ...config };
-  for (const key of ['vadModelSrc', 'projectionModelSrc', 'vocabModelSrc']) {
+  for (const key of ['vadModelSrc', 'projectionModelSrc', 'vocabModelSrc', 's3genModelSrc']) {
     if (typeof copy[key] === 'string') copy[key] = resolveSdkAsset(sdk, copy[key]);
   }
   return copy;