android/TrainingRuntimeCheck.java

import java.io.File;
import java.util.Arrays;
import java.util.HashMap;
import java.util.Map;
import java.nio.ByteBuffer;
import java.nio.ByteOrder;
import org.tensorflow.lite.Interpreter;

/** Runs under Android app_process with the matched TensorFlow Lite 2.16.1 AARs. */
public final class TrainingRuntimeCheck {
  static float[][][][] image(int h,int w) {
    float[][][][] x=new float[1][3][h][w];
    for(int c=0;c<3;c++)for(int y=0;y<h;y++)for(int k=0;k<w;k++)x[0][c][y][k]=(float)Math.sin(c+y*.07+k*.11);
    return x;
  }
  static Map<String,Object> inputs(Object... values) {
    Map<String,Object> out=new HashMap<>();for(int n=0;n<values.length;n+=2)out.put((String)values[n],values[n+1]);return out;
  }
  static float[] infer(Interpreter i,float[][][][] x,int count) {
    float[][] scores=new float[1][count];i.runSignature(inputs("x",x),inputs("scores",scores),"infer");return scores[0];
  }
  static float train(Interpreter i,float[][][][] x,float[][] y) {
    ByteBuffer rate=ByteBuffer.allocateDirect(4).order(ByteOrder.nativeOrder());rate.putFloat(0,.0001f);
    ByteBuffer loss=ByteBuffer.allocateDirect(4).order(ByteOrder.nativeOrder());
    i.runSignature(inputs("x",x,"y",y,"learning_rate",rate),inputs("loss",loss),"train");return loss.getFloat(0);
  }
  static void check(boolean okay,String message){if(!okay)throw new AssertionError(message);}
  public static void main(String[] args) {
    File model=new File(args[0]);String checkpoint=args[1];int classes=Integer.parseInt(args[2]);
    Interpreter.Options options=new Interpreter.Options().setNumThreads(2).setUseXNNPACK(false);
    float[][][][] x=image(224,224);float[][] target=new float[1][classes];target[0][7]=1;
    float[] trained,continued;float first=0,last=0;
    System.out.println("ANDROID_CHECK opening interpreter");
    try(Interpreter i=new Interpreter(model,options)) {
      System.out.println("ANDROID_CHECK interpreter open");
      float[] original=infer(i,x,classes);
      System.out.println("ANDROID_CHECK inference complete");
      for(int n=0;n<6;n++){last=train(i,n%2==0?x:image(192,320),target);if(n==0)first=last;check(Float.isFinite(last),"loss nonfinite");System.out.println("ANDROID_CHECK step="+n+" loss="+last);}
      trained=infer(i,x,classes);check(!Arrays.equals(original,trained),"weights did not alter output");
      i.runSignature(inputs("checkpoint_path",checkpoint),new HashMap<>(),"save");
      check(new File(checkpoint).length()>0,"empty checkpoint");
      train(i,x,target);continued=infer(i,x,classes);
    }
    try(Interpreter restored=new Interpreter(model,options)) {
      restored.runSignature(inputs("checkpoint_path",checkpoint),new HashMap<>(),"restore");
      check(Arrays.equals(trained,infer(restored,x,classes)),"restore differs");
      train(restored,x,target);check(Arrays.equals(continued,infer(restored,x,classes)),"optimizer resume differs");
    }
    System.out.println("ANDROID_TRAINING_PASS {\"model\":\""+model.getName()+"\",\"mixed_shapes\":[[224,224],[192,320]],\"first_loss\":"+first+",\"last_loss\":"+last+",\"restart_exact\":true,\"optimizer_resume_exact\":true,\"runtime\":\""+org.tensorflow.lite.TensorFlowLite.runtimeVersion()+"\",\"device\":\"Android emulator x86_64\"}");
  }
}