Ease main thread by updating UI on every Nth token only. (#2760)

Summary:

.

Reviewed By: kirklandsign

Differential Revision: D55532081

Author: shoumikhin

examples/demo-apps/apple_ios/LLaMA/LLaMA/Application/ContentView.swift

@@ -204,15 +204,21 @@ struct ContentView: View {
         return
       }
       do {
+        var tokens: [String] = []
         try runnerHolder.runner?.generate(text, sequenceLength: seq_len) { token in
-
-          DispatchQueue.main.async {
-            withAnimation {
-              var message = messages.removeLast()
-              message.text += token
-              message.tokenCount += 1
-              message.dateUpdated = Date()
-              messages.append(message)
+          tokens.append(token)
+          if tokens.count > 2 {
+            let text = tokens.joined()
+            let count = text.count
+            tokens = []
+            DispatchQueue.main.async {
+              withAnimation {
+                var message = messages.removeLast()
+                message.text += text
+                message.tokenCount += count
+                message.dateUpdated = Date()
+                messages.append(message)
+              }
             }
           }
           if shouldStopGenerating {

examples/demo-apps/apple_ios/LLaMA/LLaMA/Application/ResourceMonitor.swift

@@ -12,7 +12,7 @@ final class ResourceMonitor: ObservableObject {
   }
 
   public func start() {
-    memoryUpdateTimer = Timer.scheduledTimer(withTimeInterval: 0.5, repeats: true) { [weak self] _ in
+    memoryUpdateTimer = Timer.scheduledTimer(withTimeInterval: 1, repeats: true) { [weak self] _ in
       self?.updateMemoryUsage()
     }
   }