[Android] test for async invoke

Add testcase for async invoke with flexible tensor using llama.cpp.

Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>

Author: jaeyun-jung

java/android/nnstreamer/src/androidTest/assets/README.txt

@@ -7,7 +7,10 @@ $ tree .
     │   ├── config_pipeline_imgclf.conf
     │   ├── config_pipeline_imgclf_key.conf
     │   ├── config_single_imgclf.conf
-    │   └── config_single_imgclf_key.conf
+    │   ├── config_single_imgclf_key.conf
+    │   └── config_single_llamacpp_async.conf
+    ├── llamacpp_data
+    │   └── tinyllama-1.1b-chat-v1.0.Q2_K.gguf
     ├── pytorch_data
     │   ├── mobilenetv2-quant_core-nnapi.pt
     │   └── orange_float.raw
@@ -28,3 +31,23 @@ $ tree .
         ├── orange.png
         ├── orange.raw
         └── test_video.mp4
+
+
+Configuration example:
+The configuration file is a json formatted string for ML-service feature, which describes the configuration to run an inference application with model-agnostic method.
+If you implement new Android application with ML-service and need to get a model from application's internal storage, you can set the predefined entity string '@APP_DATA_PATH@'.
+Then it will be replaced with the application's data path.
+
+Below is an example of asynchronous inference using LLaMA C++.
+
+config_single_llamacpp_async.conf
+{
+    "single" :
+    {
+        "framework" : "llamacpp",
+        "model" : ["@APP_DATA_PATH@/nnstreamer/llamacpp_data/tinyllama-1.1b-chat-v1.0.Q2_K.gguf"],
+        "custom" : "num_predict:32",
+        "invoke_dynamic" : "true",
+        "invoke_async" : "true"
+    }
+}

java/android/nnstreamer/src/androidTest/java/org/nnsuite/nnstreamer/APITestMLService.java

@@ -1025,4 +1025,55 @@ public void testRunSingleShotRegistered() {
             MLService.Model.delete(name, 0);
         }
     }
+
+    @Test
+    public void testRunAsyncInvoke() {
+        if (!NNStreamer.isAvailable(NNStreamer.NNFWType.LLAMACPP)) {
+            /* cannot run the test */
+            return;
+        }
+
+        String config = APITestCommon.getConfigPath() + "/config_single_llamacpp_async.conf";
+
+        try {
+            MLService.EventListener asyncEventListener = new MLService.EventListener() {
+                @Override
+                public void onNewDataReceived(String name, TensorsData data) {
+                    if (data == null || data.getTensorsCount() != 1) {
+                        mInvalidState = true;
+                        return;
+                    }
+
+                    mReceived++;
+                }
+            };
+
+            MLService service = new MLService(config, asyncEventListener);
+
+            service.start();
+
+            /* push input buffer */
+            String inputText = "Hello my name is";
+            ByteBuffer buffer = TensorsData.allocateByteBuffer(inputText);
+
+            TensorsInfo info = service.getInputInformation(null);
+            assertEquals(NNStreamer.TensorFormat.FLEXIBLE, info.getFormat());
+
+            TensorsData input = TensorsData.allocate(info);
+            input.setTensorData(0, buffer);
+
+            service.inputData(null, input);
+
+            /* sleep 3 seconds to invoke */
+            Thread.sleep(3000);
+
+            /* check received data from output node */
+            assertFalse(mInvalidState);
+            assertTrue(mReceived > 1);
+
+            service.close();
+        } catch (Exception e) {
+            fail();
+        }
+    }
 }