Backport MacOS SDK fix from main

.github/workflows/release.yaml

@@ -8,7 +8,7 @@ on:
 jobs:
   # Full build of the Mac assets
   build-darwin:
-    runs-on: macos-latest
+    runs-on: macos-12
     environment: release
     steps:
       - uses: actions/checkout@v4
@@ -38,6 +38,8 @@ jobs:
           APPLE_PASSWORD: ${{ secrets.APPLE_PASSWORD }}
           APPLE_TEAM_ID: ${{ vars.APPLE_TEAM_ID }}
           APPLE_ID: ${{ vars.APPLE_ID }}
+          SDKROOT: /Applications/Xcode_13.4.1.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk
+          DEVELOPER_DIR: /Applications/Xcode_13.4.1.app/Contents/Developer
         run: |
           ./scripts/build_darwin.sh
         

llm/ext_server/server.cpp

@@ -1007,13 +1007,15 @@ struct llama_server_context
                 slot.n_sent_text += result.text_to_send.size();
                 // add the token to slot queue and cache
             }
-            slot.add_token_string(result);
+
             if (slot.params.stream)
             {
                 send_partial_response(slot, result);
             }
         }
 
+        slot.add_token_string(result);
+
         if (incomplete)
         {
             slot.has_next_token = true;