tf.random.truncated_normal([num_classes, n_hidden_3],
                                           stddev=0.1,
                                           seed=seed)),
    }

    self.biases = {
        'b1': tf.Variable(tf.zeros([n_hidden_1])),
        'b2': tf.Variable(tf.zeros([n_hidden_2])),
        'b3': tf.Variable(tf.zeros([n_hidden_3])),
        'b4': tf.Variable(tf.zeros([num_classes])),
    }

  @tf.function

  size_t start_map = fbb.StartMap();

  if (failed(AddIntAttr(func, attrs, "max_detections", &fbb)) ||
      failed(AddIntAttr(func, attrs, "max_classes_per_detection", &fbb)) ||
      failed(AddIntAttr(func, attrs, "num_classes", &fbb)) ||
      failed(AddFloatAttr(func, attrs, "nms_score_threshold", &fbb)) ||
      failed(AddFloatAttr(func, attrs, "nms_iou_threshold", &fbb)) ||
      failed(AddFloatAttr(func, attrs, "y_scale", &fbb)) ||

    "type: 'int'} attr : { name: 'max_detections' type: 'int'} attr : { "
    "name: 'nms_iou_threshold' type: 'float'} attr : { name: "
    "'nms_score_threshold' type: 'float'} attr : { name: 'num_classes' type: "
    "'int'} attr : { name: 'w_scale' type: 'float'} attr : { name: 'x_scale' "
    "type: 'float'} attr : { name: 'y_scale' type: 'float'} attr { name: "

  %0 = "tf.Const"() {value...

    Arg<TF_FloatTensor, [{batch_size x num_classes matrix}]>:$features,
    Arg<TF_I32OrI64Tensor, [{batch_size vector with values in [0, num_classes).
This is the label for the given minibatch entry.}]>:$labels
  );

  let results = (outs
    Res<TF_FloatTensor, [{Per example loss (batch_size vector).}]>:$loss,
    Res<TF_FloatTensor, [{backpropagated gradients (batch_size x num_classes matrix).}]>:$backprop
  );

                }

                IndexBuilder rootBuilder = null;
                try {
                    int numClasses = input.readInt(true);
                    rootBuilder = new IndexBuilder();

                    for (int classCounter = 0; classCounter < numClasses; ++classCounter) {
                        long classId = input.readLong(true);

            File dstDir = new File(workspace, OUT_CLASSES);
            return new GeneratedAccessors(srcDir, dstDir);
        }

        @Override
        public void visitOutputs(File workspace, OutputVisitor visitor) {
            visitOutputDir(visitor, workspace, OUT_SOURCES);
            visitOutputDir(visitor, workspace, OUT_CLASSES);
        }

quantized_op = true, detections_per_class = 100 : i64, device = "", h_scale = 5.000000e+00 : f32, max_classes_per_detection = 1 : i64, max_detections = 20 : i64, nms_iou_threshold = 6.000000e-01 : f32, nms_score_threshold = 3.000000e-01 : f32, num_classes = 90 : i64, use_regular_nms = false, w_scale = 5.000000e+00 : f32, x_scale = 1.000000e+01 : f32, y_scale = 1.000000e+01 : f32} : (tensor<1x64x64x32xf32>, tensor<1x64x64x32xf32>, tensor<1x64x64x32xf32>) -> (tensor<1x32x32x32xf32>, tensor<1x32x32x32xf32>,...