Keep some leeway for fs.stat

desktop/src/main/services/ml-worker.ts

@@ -184,14 +184,13 @@ const downloadModel = async (saveLocation: string, name: string) => {
 /**
  * Create an ONNX {@link InferenceSession} with some defaults.
  */
-const createInferenceSession = async (modelPath: string) => {
-    return await ort.InferenceSession.create(modelPath, {
+const createInferenceSession = async (modelPath: string) =>
+    ort.InferenceSession.create(modelPath, {
         // Restrict the number of threads to 1.
         intraOpNumThreads: 1,
         // Be more conservative with RAM usage.
         enableCpuMemArena: false,
     });
-};
 
 const cachedCLIPImageSession = makeCachedInferenceSession(
     "mobileclip_s2_image_opset18_rgba_opt.onnx",
@@ -233,9 +232,11 @@ const getTokenizer = () => (_tokenizer ??= new Tokenizer());
 export const computeCLIPTextEmbeddingIfAvailable = async (text: string) => {
     const sessionOrSkip = await Promise.race([
         cachedCLIPTextSession(),
-        // Wait for a tick to get the session promise to resolved the first time
-        // this code runs on each app start (and the model has been downloaded).
-        wait(0).then(() => 1),
+        // Wait a bit to get the session promise to resolved the first time this
+        // code runs on each app start (in these cases the model will already be
+        // downloaded, so session creation should take only a 1 or 2 ticks: file
+        // system stat, and ort.InferenceSession.create).
+        wait(50).then(() => 1),
     ]);
 
     // Don't wait for the download to complete.