can it work under tf 2.x?

[lg2578]

1.get_gradients only support in tf 1 lengcy graph mode ,the code seems can't work under tf 2.x
2.I've runned the fashion-mnist demo, the model's compile method not init the model's weight, so may need to create model again before run the gctf.
like the following italic code:
import gctf #import gctf

time_callback_gctf = TimeHistory()
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(512, activation=tf.nn.relu),
tf.keras.layers.Dense(256, activation=tf.nn.relu),
tf.keras.layers.Dense(64, activation=tf.nn.relu),
tf.keras.layers.Dense(512, activation=tf.nn.relu),
tf.keras.layers.Dense(256, activation=tf.nn.relu),
tf.keras.layers.Dense(64, activation=tf.nn.relu),
tf.keras.layers.Dense(10, activation=tf.nn.softmax)])

model.compile(optimizer = gctf.optimizers.adam(),
loss = 'sparse_categorical_crossentropy',
metrics=['acc'])

[lg2578]

with a random seed =3, I run the fashion-mnist demo,the ruslt is:
Type Execution time Accuracy Loss

---

Model without gctf: 28.0661 0.886354 0.310736
Model with gctf 29.7088 0.884479 0.319094

tf version is 2.5.1
why it happens?